Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
GIS analysis of helicopter rescue in San Bernardino County, California
(USC Thesis Other)
GIS analysis of helicopter rescue in San Bernardino County, California
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
GIS Analysis of Helicopter Rescue in San Bernardino County, California
by
Daniel R. Fisher
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)
December 2021
Copyright © 2021 Daniel R. Fisher
ii
Dedication
To my parents, Wayne and Natalie Fisher; my wife, Brianna Crawley; and my children, Roscoe,
Avelina, and Merielle.
iii
Acknowledgements
I would like to thank my advisor, Dr. Jennifer Bernstein, for her guidance throughout this
project, and the faculty and staff of the Spatial Sciences Institute for the GIS Master’s program.
Every part of my experience, from the application and registration processes to the design of the
core and elective courses, the technical support for the software, and the thesis project have been
excellent.
Thanks to Air Rescue team members Peter Peng, Melanie Munns, and Steve Miller for
creating the rescue spreadsheet and transcribing the hoist log books. Steve Miller also provided
immeasurable help to this thesis by sharing his knowledge of Air Rescue operations and
providing comments on a draft.
Thanks to Robert O’Brine, the captain at the time of writing of the San Bernardino
Sheriff’s Emergency Operations Division for providing comments on a draft of this thesis, state
and county code sections, and images of aviation unit helicopters.
Thanks to my parents, Wayne and Natalie Fisher, who provided childcare and took over
home school during my thesis writing process.
Finally, thanks to my wife, Brianna Crawley, for her patience and support throughout my
time completing this program.
iv
Table of Contents
Dedication ....................................................................................................................................... ii
Acknowledgements ........................................................................................................................ iii
List of Figures ................................................................................................................................ vi
List of Abbreviations .................................................................................................................... vii
Abstract ........................................................................................................................................ viii
Chapter 1 Introduction .................................................................................................................... 1
1.1. The San Bernardino County Sheriff’s Air Rescue Program ...............................................1
1.2. Helicopter Hoist Rescue .....................................................................................................4
1.3. Sheriff’s Aviation Aircraft and Equipment.........................................................................5
1.3.1 Patrol Helicopters ...................................................................................................5
1.3.2 Medium Helicopters ...............................................................................................5
1.3.3 Rescue Devices ......................................................................................................7
1.4. Dispatch Protocols and Decision Making ...........................................................................8
1.5. Helicopter Capabilities and Density Altitude .....................................................................9
1.6. GIS Analysis of Rescue Records ......................................................................................12
Chapter 2 Related Work................................................................................................................ 13
2.1. Georeferencing Described Locations in Historical Records .............................................13
2.2. Statistical Spatial Analysis of Incident Data .....................................................................16
2.3. Spatial Analysis of Search and Rescue Incidents .............................................................19
Chapter 3 Methodology ................................................................................................................ 24
3.1. Rescue Geodatabase..........................................................................................................24
3.1.1. Sheriff’s Aviation Documentation ...........................................................................24
3.1.2. Database Construction .............................................................................................29
3.1.3. Selecting Hiking and Climbing Records..................................................................34
v
3.2. Spatial Analysis of Rescues ..............................................................................................35
3.2.1. Hexagonal Grid and Visualization of Rescues by Elevation ...................................36
3.2.2. Moran’s I Calculation ..............................................................................................39
3.2.3. Getis-Ord Gi
*
Calculation ........................................................................................41
3.2.4. Temporal Density and Hot Spot Calculations .........................................................42
3.3. Kernel Density ..................................................................................................................42
Chapter 4 Results .......................................................................................................................... 45
4.1. Trends in the Elevation of Hoist Rescues .........................................................................45
4.2. Statistical and Kernel Density Calculations ......................................................................48
4.3. Monthly Hot Spots and Kernel Densities of Rescue Incidents .........................................53
Chapter 5 Conclusion .................................................................................................................... 62
5.1. Trends in Hoist Rescues....................................................................................................62
5.2. Hoist Rescue Database Analysis Possibilities ..................................................................64
5.2.1. Rescue Database Analysis Possibilities ...................................................................64
5.2.2. Application Dashboard ............................................................................................65
5.2.3. Information Gathering .............................................................................................66
5.3. Other Applications of GIS for Air Rescue ........................................................................68
5.4. GIS Analysis of Rescue Operations ..................................................................................70
References ..................................................................................................................................... 71
vi
List of Figures
Figure 1 San Bernardino County, Sheriff’s Aviation, and Trauma Centers ................................... 2
Figure 2 40 King 3 conducting a hoist ............................................................................................ 6
Figure 3 Air Rescue 306 conducting a hoist with a Stokes litter .................................................... 7
Figure 4 Density Altitude Chart .................................................................................................... 11
Figure 5 A page of a hoist log book .............................................................................................. 25
Figure 6 Format of the Excel hoist rescue spreadsheet................................................................. 26
Figure 7 Most recent Blue Card format ........................................................................................ 29
Figure 8 Georeferencing process for a rescue in Cucamonga Canyon ......................................... 32
Figure 9 Rescue location polygons and hexagonal grid ............................................................... 38
Figure 10 Quartic Kernel Function ............................................................................................... 43
Figure 11 Hoist Rescues above and below 6,000 ft ...................................................................... 46
Figure 12 Statistics of rescues above and below 6,000 ft ............................................................. 47
Figure 13 Rescue kernel density and Gi
*
values ........................................................................... 50
Figure 14 Kernel densities and hot spots for March, April, and May .......................................... 54
Figure 15 Kernel densities and hot spots for June, July, and August ........................................... 56
Figure 16 Kernel densities and hot spots for September, October, and November ...................... 58
Figure 17 Kernel densities and hot spots for December, January, and February ......................... 60
vii
List of Abbreviations
DEM Digital Elevation Model
GIS Geographic information science
GPS Global Positioning System
HLZ Helicopter Landing Zone
MAUP Modifiable Areal Unit Problem
NVG Night Vision Goggles
SAR Search and Rescue
SBSD San Bernardino County Sheriff’s Department
SSI Spatial Sciences Institute
TAZ Traffic Analysis Zone
USC University of Southern California
USCG United States Coast Guard
viii
Abstract
People become lost and injured in remote areas on a daily basis. Search and rescue personnel,
including members of the military, law enforcement agencies, fire departments, emergency
medical services, and volunteer teams stand by for the opportunity to rescue someone. This
requires the development of teamwork and skills and the acquisition and maintenance of
equipment. Rapid and accurate analysis of geographic information plays a critical role in
ensuring resources are used as safely and effectively as possible in search and rescue operations.
Historical search and rescue records contain a wealth of information that can guide
training, drive recruitment of new members, and define the necessary capabilities of equipment
and personnel. Although they may not have been collected with the intent of providing data for
spatial analysis, the geographic information contained in rescue documentation can provide
information as to the trends in where rescues are occurring, which can help anticipate future
incidents and improve readiness.
This thesis is a retrospective application of geographic information science to the records
of the San Bernardino County sheriff’s air rescue program. The hoist rescues performed since the
program’s inception in 1993 were georeferenced, digitized, and used as the basis for a
geodatabase. The spatial pattern of air rescues in the county was analyzed to identify the areas of
greatest rescue activity and seasonal trends in locations and elevations. This information will
help leaders on the team guide training and preparation and help define the necessary capabilities
of rescue helicopters for work in the county. In addition, by showing trends in rescue activity in
certain areas and times of the year, recommendations are made for preventative measures to keep
hikers and climbers safe. The analysis methods here can be applied to similar projects throughout
the United States and beyond.
1
Chapter 1 Introduction
The San Bernardino County sheriff’s air rescue program provides helicopter search and rescue
and medical transport services throughout the county. Search and rescue is an inherently
geospatial activity, as location and terrain dictate every aspect of training for, planning,
managing, and executing missions. Rotary-wing search and rescue is particularly demanding for
the helicopter and aircrew and depends strongly on teamwork, elevation, and weather conditions.
This project will use geographic information science (GIS) to analyze the activities and enhance
the capabilities of the sheriff’s air rescue program by creating a database of rescues and
analyzing trends in the data.
1.1. The San Bernardino County Sheriff’s Air Rescue Program
The county of San Bernardino, California, with an area of over 20,000 square miles, is
the largest county in the continental United States. It is approximately the size of the state of
West Virginia. The San Bernardino Mountains cut across the southwestern part of the county,
and the Mojave Desert occupies the area north and east of the mountains. A section of Joshua
Tree National Park is in the eastern part of the county. Approximately eighty percent of the
population lives south of the mountains in the Ontario-San Bernardino-Riverside metropolitan
area (San Bernardino County 2015). Figure 1 shows the state of California, San Bernardino
County, and the locations of the sheriff’s aviation helicopters and trauma centers with helipads.
The terrain in the county is extremely varied. The San Bernardino Mountains include
several peaks above 10,000 feet, with the tallest, San Gorgonio Mountain, at 11,503 feet. The
mountains receive abundant snow in the winter, while it may be hot and dry in the Mojave
Desert at the same time.
2
Figure 1 San Bernardino County, sheriff’s aviation, and trauma centers
3
Every year, hundreds of people become lost or injured in remote areas of San Bernardino
County. There are many highly trafficked hiking and climbing areas in the mountains, with
trailheads easily accessible from all of southern California. The desert areas north of the
mountains have hiking trails and many areas for riding dirt bikes and other off-road vehicles.
Section 26614 of the California Government Code states that “[t]he board of supervisors of a
county may authorize the sheriff to search for and rescue persons who are lost or are in danger of
their lives within or in the immediate vicinity of the county.” (California Legislative Information
2020) Section 12.0511 of the San Bernardino County Code of Ordinances makes this
authorization and states that “the Sheriff shall have the authority to search for and rescue persons
who are lost or are in danger of their lives within or in the immediate vicinity of the County.”
(San Bernardino County, CA Code of Ordinances 2019) In accordance with these statutes, the
sheriff’s department coordinates volunteer search and rescue teams that respond with trained
personnel to find, rescue, treat, and evacuate lost and injured people in remote areas (San
Bernardino County Sheriff’s SAR 2020).
The sheriff’s department also maintains a rescue helicopter, which provides 24/7 search
and rescue and medical transport services throughout the county. The sheriff’s air rescue team is
made up of physicians, physician assistants, nurses, paramedics, and technical rescue experts
who volunteer their time on the rescue helicopter (San Bernardino County Sheriff’s Department
2020). The volunteer team provides medics on Friday, Saturday, and Sunday, while San
Bernardino County Fire paramedics staff the rescue helicopter Monday through Thursday. Full-
time sheriff’s deputies in the aviation unit are also trained as rescuers and perform rescues after
hours. The rescue helicopters have medical equipment comparable to an ambulance, although,
depending on their qualifications, some air medics have a greater scope of practice than a
4
paramedic on an ambulance or fire engine. The vast size, long distances, and rugged terrain of
the county mean that rescuers using helicopters can often find, access, treat, and transport
patients more quickly than ground resources.
1.2. Helicopter Hoist Rescue
The safest way to get a person into or out of a helicopter is for it to land. The sheriff’s
department uses two procedures to insert or extract people if it is unable to land. If the helicopter
can hover within a few feet of the ground, rescuers can do a hover step, in which they enter or
exit by climbing the skids. If trees or terrain prevent a hover step, a hoist is necessary. A hoist
involves lowering the rescuer and equipment to the ground using a winch with a steel cable while
the helicopter hovers. The rescuer packages the victim in a harness or litter, then the victim and
rescuer are raised back into the helicopter. This type of rescue is demanding for the helicopter
and aircrew. The ability to perform a hoist rescue depends on a combination of the power and
weight of the helicopter; the amount of fuel, personnel, and equipment on board; the skill and
teamwork of the pilot, crew chief, and medic; and the elevation, temperature, and wind.
When performing a hoist rescue, the sheriff’s department helicopters usually fly with one
pilot, one crew chief, and one or two air medics. The crew chief has the greatest responsibility
during a hoist operation. While the helicopter is flying to the target area, the crew chief ensures
the cabin is ready and reviews emergency procedures with the pilot. When the helicopter is near
the target area, the crew chief opens the door and stands on the skid. At this point, the pilot
cannot see the ground below the helicopter, so the crew chief guides the pilot into position using
the intercom. When the helicopter is hovering in the right place, the crew chief ensures the hook
is connected properly and operates the hoist to lower the rescuer and any necessary equipment to
the ground. When the rescuers have packaged the victim, the crew chief raises the victim and
5
rescuers back into the helicopter. The team almost always has one person at a time on the hoist,
so multiple evolutions are necessary to recover more than one person.
1.3. Sheriff’s Aviation Aircraft and Equipment
The San Bernardino County sheriff’s aviation unit has a fleet of helicopters and airplanes.
Its main facility is at the San Bernardino International Airport, with one patrol helicopter on duty
at the Apple Valley Airport.
1.3.1 Patrol Helicopters
The sheriff’s patrol helicopters are Eurocopter AS350 B3 “AStar” or Airbus H125
models. They are almost identical but have different names because Eurocopter became Airbus
Helicopters in 2014 (Faury 2014). They are single-engine, three-blade light utility helicopters
with 848 horsepower and 3433 lb. empty weight (Airbus 2020). They have a cargo hook that can
carry a bucket for fire-fighting operations. They can also be equipped with an external hoist for
rescues. The hoist is a Goodrich 44301-10-5 Rescue Hoist with 170 feet of 3/16 inch steel cable
and a maximum rated load of 500 lbs (Goodrich Hoist and Winch 2020). The small size of the
cabin usually prevents victims from being brought inside while the helicopter is in flight, so they
are kept outside on the hook until it can land, at which time they are brought inside or transferred
to a ground ambulance. The call sign for the patrol helicopters is 40 King followed by the last
digit of the tail number. Figure 2 shows 40 King 3 during a hoist operation with the crew chief
standing on the skid.
1.3.2 Medium Helicopters
The sheriff’s department has three Huey-type helicopters that are mostly used for rescue.
They can also be equipped with buckets or tanks for firefighting. Air Rescue 306 is a Super
6
Figure 2 40 King 3 conducting a hoist.
Huey, which means that it has an upgraded engine, transmission, drive shaft, and tail rotor
compared to a regular Huey. It has one engine, approximately 1800 horsepower, and 5721 lb.
empty weight. Air Rescue 307 is a Bell 212, which is the civilian version of the military Huey. It
has two engines, approximately the same horsepower as 306, and 6710 lb. empty weight. Air
Rescue 308 is a regular Huey, with one engine, 1300 horsepower, and 5338 lb. empty weight.
The medium helicopters can be equipped with a Lucas Aerospace rescue hoist, which has been
known as the Goodrich 43205 since the Goodrich Corporation acquired the successor of Lucas
Aerospace in 2002 (Goodrich Hoist and Winch 2020). It has 250 feet of 3/16 inch steel cable and
a maximum rated load of 600 lbs (Goodrich Hoist and Winch 2020). Figure 3 shows Air Rescue
306 conducting a hoist with a Stokes litter. The medium helicopters are large enough to bring a
victim inside while in flight, where they can be treated by a medic while flying to the hospital.
7
For both the patrol and medium helicopters, the hoist creates an off-center load on the
helicopter, which the pilot must counter to maintain a stable hover. In most cases, the ability of
the pilot to correct for the changing center of gravity limits the amount of weight that can be
lifted before the rated capacity of the hoist is reached.
Figure 3 Air Rescue 306 conducting a hoist with a Stokes litter.
1.3.3 Rescue Devices
For hoist rescues of ambulatory patients, the air medics use the Hotseat, which is a nylon
harness manufactured by CMC. It is put on like a jacket and has rings that connect to the hoist
hook (CMC 2020).
If a victim is in such a precarious position that the rescuer is not able to disconnect from
the hoist hook, the air medics use the Lifesaver Victim Harness, which is made by CMC (CMC
2020). The victim harness is a belt with leg loops, which the rescuer puts on the victim, and the
rescuer and victim are raised up to the helicopter together.
The air medics have two devices for non-ambulatory patients. On the medium
helicopters, they use a Stokes litter, which is made from titanium or steel rails and wire mesh
8
(CMC 2020). It is designed to protect the patient but not catch the rotor wash underneath the
helicopter, which can cause it to spin. The patient is usually packaged on a rigid backboard,
which fits into the Stokes litter. The Stokes litter has a steel cable “spider strap” system, which
connects to its rails and has a ring for the hoist hook.
The patrol helicopters do not have enough room inside for a rigid Stokes litter. For non-
ambulatory patients, they use the Bauman Air Bag, which is a soft nylon litter. The patient is
packaged on a rigid backboard, then placed into the bag and secured with straps. The Bauman
Air Bag has nylon straps that connect to a ring for the hoist hook.
1.4. Dispatch Protocols and Decision Making
The rescue helicopter is usually dispatched for rescues by the San Bernardino County
Consolidated Fire Agencies (CONFIRE) dispatch center, which dispatches fire and emergency
medical services and coordinates mutual aid requests throughout the county (San Bernardino
County Fire 2020). The dispatchers can dispatch the rescue helicopter if they receive a call in
which the reporting party describes a victim in a place that will be difficult or impossible for
ground resources to access. The dispatchers can instruct callers in remote areas with GPS-
enabled phones to access their location information and read their coordinates, which they can
include with the dispatch information.
The rescue helicopter is also often requested by ground units who are unable to reach a
victim or make contact and face a protracted extraction to the road due to steep terrain, water, or
other hazards. On mountain roads, which are often narrow and windy with steep drop offs and no
guard rails, cars that go off the road may not be far from it, but the terrain may be so steep and
unstable they are difficult to access and require a hoist rescue. Even in populated areas, there can
9
be situations where victims are inaccessible, such as when people become stranded in flood
control channels.
Ground units can request the helicopter if their driving time to the hospital is longer than
the victim’s condition will allow. The rescue helicopter responds to motor vehicle collisions with
severe injuries in remote parts of the county because its transport time to the hospital can be
much faster than that of a ground ambulance. In these cases, the helicopter is usually able to land
on the road near the site of the incident and a hoist rescue is not necessary.
Decisions taking into account expense, training of personnel, maintenance requirements,
and weight have to be made and continually revised about the equipment the helicopter carries
and the readiness posture of the team. The records of rescues the sheriff’s department has
performed can give a basis to these decisions and also guide efforts to prevent injuries from
happening. This project will focus on the spatial and temporal patterns of hoist rescues of hikers
and climbers. Hoist rescues also often involve operators of on- and off-road vehicles, but because
they travel faster and cover longer distances, they tend to occur in extremely remote areas and
have a very diffuse spatial distribution. The hoist rescues of hikers and climbers occur in more
concentrated areas, which will be more amenable to preventative measures.
1.5. Helicopter Capabilities and Density Altitude
To conduct a hoist rescue, a helicopter must maintain a stable hover and take on an
external load. This is a demanding task, because hovering requires more power than landing or
forward flight. The helicopter’s ability to hover depends on a number of factors, the most
important of which are its total weight, the power of its engine, and the ability of its tail rotor to
counter the torque of the main rotor blades.
10
A helicopter requires more power to hover at high altitudes because there is less air
pressure for the rotor blades to push against and generate lift. High temperatures and humidity
are unfavorable because they have the effect of lowering the air pressure. Altitude and
temperature can be combined in a formula to approximate the density altitude, which is the
altitude the helicopter “feels” it is flying (AOPA 2020):
𝐴 = 𝐴 + [120 × (𝑇 − 𝑇 )], (1)
Where 𝐴 is the density altitude in feet, 𝐴 is the uncorrected altitude, 𝑇 is the outside air
temperature in degrees Celcius, and 𝑇 = 15 − × , the standard temperature at the
uncorrected altitude.
Using the formula, the density altitude at 6,000 ft. and 90
0
F (32
0
C) is approximately
9,500 ft. Figure 4 is a chart that allows the user to calculate density altitude based on pressure
altitude and temperature (Schappert 2020).
Air Rescue 308 is a Huey and is the least capable helicopter in the sheriff’s department’s
fleet. It was recently equipped with a hoist and has been used to perform several hoist rescues. It
is the assessment of the pilots in the aviation unit that Air Rescue 308 can perform hoist rescues
at elevations up to 6000 ft. Above this elevation the helicopter is not able to reliably perform
hoists.
Mt. San Antonio and San Gorgonio Mountain, two of the most popular mountains for
hiking in San Bernardino County, have elevations of 10,069 and 11,503 ft., respectively. They
see heavy traffic and rescue activity year round. Big Bear Lake, which is surrounded by
mountains with ski areas, hiking trails, and rock climbing locations, has an elevation of 6,750 ft.
With so many possibly hazardous outdoor activities occurring above 6,000 ft., it is reasonable to
expect there will be a significant need for hoist rescues at and above that elevation.
11
The database of rescues the aviation unit has performed will allow the spatial distribution
of rescues above and below 6000 ft. to be visualized and demonstrate the importance of a
helicopter to be capable of hoist rescues above this level. These results will help the leadership of
the aviation unit to determine the role of Air Rescue 308 as a rescue ship and the necessary
capabilities of a new rescue helicopter, if it is to be replaced.
Figure 4 Density Altitude Chart
12
1.6. GIS Analysis of Rescue Records
The San Bernardino sheriff’s rescue helicopter program has performed hundreds of hoist
rescues since it developed the capability in 1993. The records of these rescues were only recently
tabulated and digitized, and have not been georeferenced, mapped, or analyzed in any way. The
rescue records contain a wealth of information that can be used to preserve the legacy of the
people involved in the rescues; express the accomplishments of the program to the leadership of
the county, the sheriff’s department, and the public; understand the utilization of equipment;
guide training; and identify hazardous areas that may be candidates for preventative measures.
GIS provides the tools to organize, store, visualize, and learn from geographic
information. This project applies GIS these rescue records and demonstrates a few types of
analysis that can be performed on them. The next chapter discusses articles on georeferencing
locations from historical text descriptions, spatial analysis of search and rescue records, and
statistical spatial calculations on incident data. Chapter 3 describes the format of the rescue
records, the georeferencing process, and the details of the calculations performed. Chapter 4
presents the results of the analysis, with maps and tables showing the locations of rescues and
trends in number and location over the months of the year and the years the program has been
active. Finally, chapter 5 concludes with additional ways GIS can be applied to the rescue
helicopter program and a framework for improving data collection to facilitate future studies.
13
Chapter 2 Related Work
GIS is a framework for collecting, managing, analyzing, and visualizing information associated
with locations (Esri 2020). It provides the tools to analyze and learn from the spatial patterns of
interesting phenomena. The articles discussed in this chapter develop techniques and apply them
to topics similar to those of this project.
Section 2.1 describes publications that discuss georeferencing historical records so they
can be analyzed using GIS, and section 2.2 describes statistical spatial analysis of incident data.
Section 2.3 describes GIS analyses of search and rescue incidents to predict future demand for
services, guide training and allocate resources.
2.1. Georeferencing Described Locations in Historical Records
The primary challenge in the development of the hoist rescue database was that the
majority of the records contain a description of the location of the rescue, while only a few have
geographic coordinates. To populate the database and perform analysis, the described locations
had to be georeferenced, or assigned digital representations. Wieczorek et al. (2004) addressed a
similar problem with the millions of geological, biological, and cultural specimens in the
collections of natural history museums. Global Positioning Systems (GPS) enable modern
expeditions to record coordinates when collecting specimens, but there is also a large catalog of
legacy specimens whose locations were recorded with descriptive text (Wieczorek, Guo, and
Hijmans 2004). To make the records describing the collection of these specimens available for
spatial analysis, the locations, which are described following conventions that differ between
disciplines and institutions, have inconsistent accuracy, no standard format, and usually no
metadata describing how they were determined, must be georeferenced.
14
Wieczorek et al. describe four methods to interpret described locations and assign a digital
representation to them. The first is the point method, in which each location is given a single
point. This method ignores the fact that locality descriptions usually refer to areas, and the
resulting points do not reflect the specificity with which they were recorded, which limits their
usefulness for spatial analysis. The second is the shape method, in which the locations are
represented by polygons or buffered points or polylines. This method gives the best
representation of the location, but involves a complicated process to individually generate each
vector object and requires access to GIS software, expertise in using it, and more storage space
than a single pair of coordinates.
The third method is the bounding-box method, in which each location is described by two
coordinate pairs that form the corners of a rectangle. This method is simpler and requires less
storage space than the shape method, but results in a less specific representation of the location.
The fourth method, and the focus of the paper, is the point-radius method, in which each location
is represented by a coordinate pair and a distance that reflects the extent of its area and its
uncertainty, taking into account contributions from the locality description, map scale, datum,
and the precision and accuracy of the source material. The resulting representations are
consistent, reproducible, and provide a measure of their uncertainty, which allows them to be
used for spatial analysis.
In a later article, Doherty and Wieczorek et al. applied the shape and point-radius
methods to locations described in records of search and rescue incidents in Yosemite National
Park (Doherty, Liu, Doke, Guo, and Wieczorek 2011). They georeferenced six incidents using
both methods and found the shape method reduced the uncertainty area up to 99.2% compared to
the point-radius method, but that the time required to process an incident increased from five to
15
15 minutes for the point-radius method to 15 to 90 minutes for the shape method. The location
with the greatest percentage difference between the point-radius and shape method areas was
described as being “on the Four-Mile trail approximately 1.5 miles up from the Four-Mile
Trailhead.” The shape method allowed the location to be represented as a buffered line, while the
circle created by the point-radius method had to have a much larger area to be sure to include the
described location.
Although the authors found that the shape method resulted in more precise
representations of described locations than the point-radius method, the time it took to
georeference the records made the shape method impractical for use on a large dataset. For the
analysis part of their study, they examined 1,356 search and rescue incidents that occurred in
Yosemite National Park between 2005 and 2010 and found that 1,271, or 93%, of them could be
georeferenced using the point-radius method with a mean uncertainty radius of 560
+
/- 51 m and
uncertainty area of 3.60
+
/- 0.840 km
2
. The 95
th
percentile of the uncertainty radii was 2,026 m,
so a 2 km grid was defined over the study area. The size of the grid was chosen based on the
uncertainty of the location data, so that although the authors did not know exactly where the
rescues occurred, they could be certain they were assigned to the correct grid cell. The incidents
were aggregated over the grid and Moran’s I and Getis-Ord Gi* calculations were carried out.
The Moran’s I calculation indicated the distribution of rescues exhibited statistically significant
clustering.
The Getis-Ord Gi
*
analysis found statistically significant hot spots, mostly around the
hiking trails in the Yosemite Valley, but the uncertainty of the georeferenced locations limited
their analysis to a very large grid. A grid square with 2 km is too large an area for the result to be
useful in learning much about rescue incidents involving hikers or climbers. To conduct a higher
16
resolution analysis that would give results closer to the scale of a hiking trail and differentiate
between smaller features, the uncertainty of the underlying locations would have to be greatly
reduced. Partly based on the results of this study, the Parks Service started to use GPS units to
record coordinates during search and rescue operations and include them with the documentation
in a new records management system.
2.2. Statistical Spatial Analysis of Incident Data
In order to perform statistical spatial calculations on incident data, the study area must be
partitioned into zones, and statistics must be calculated based on the incidents occurring in each
zone. The zones provide the foundation for statistical calculations, which can determine if a
spatial pattern exhibits statistically significant clustering or is the result of a random process. The
amount of memory and computer time required to perform calculations increase with the number
of zones. For phenomena with a high degree of spatial heterogeneity, a partitioning scheme with
identical zones, such as a regular grid, may be undesirable because there may be a small number
of cells with significant values separated by large numbers of cells with insignificant values. To
avoid wasting memory and computer time on areas that will not contribute to the results,
researchers partition their study areas into zones based on their data and study areas. The details
of the partitioning scheme must be carefully considered, because the results of the statistical
calculations depend on the relationships between the zones. This phenomenon is known as the
modifiable areal unit problem (MAUP) (Openshaw 1983). The references in this section perform
statistical calculations on incident data. Although they do not address wilderness search and
rescue incidents, the techniques can be applied to any point pattern.
A journal article by Xu et al. considers the effects of different aggregation schemes on the
analysis of traffic collisions, a phenomenon with extreme spatial heterogeneity (Xu, Huang,
17
Donga, and Abdel-Aty 2014). To assess the crash risk of roads, regional safety modelers
correlate the number of crashes with environmental factors such as the amount of traffic, density
of intersections, speed limit, and land use, as well as demographic factors such as population
density, median household income, and employment status. To perform a sensitivity analysis on
the aggregation scheme, the authors start with small base units and explore how the results of
statistical calculations change when they are combined into larger regions
The authors used 738 traffic analysis zones (TAZ) in Hillsborough County, Florida as
their base units. TAZs are used in transportation planning and are not specifically made for
traffic crash analysis. They partitioned the TAZs based on the homogeneity of their risk factors
according to 14 different schemes that had between 50 and 700 resulting zones. They found that
the spatial autocorrelation of the number of crashes calculated by the Moran’s I statistic
decreased with the number of zones, possibly because areas with similar crash risk were
aggregated together. The authors explored how the sensitivity of the variables in their model
changed with aggregation and found that aggregations with smaller units had more significant
variables. The correlation of median household income and population density with crashes were
consistent over variations in aggregation. Overall, the authors found their results were consistent
with the original TAZs and they could avoid the MAUP with aggregations that had
approximately 350 zones, about half the original number.
In a similar study, Nelson and Brewer compare the spatial autocorrelation of median
income for the state of Pennsylvania and cancer diagnosis rates for the state of New York across
census block, census tract, and county levels of aggregation (Nelson and Brewer 2015). To
understand the stability of the values across the different scales, they calculated the deviations of
the median incomes and cancer diagnosis rates for the counties from the census tracts and the
18
counties and census tracts from the census blocks within them. They found that the spatial
clustering of cancer diagnosis rates was less significant than that of median income, but they had
similar trends across scales. They found the smaller geographical units had greater variability
and that the relationships between census blocks and tracts were more similar than between
census blocks and counties.
They conclude that there is not one level of aggregation that is correct in every case, but
the appropriate level of aggregation depends on the stability of the phenomenon being studied
across different scales, and that exploring the effects of aggregation can increase the
understanding of phenomena by showing the scale of their operations and interactions.
A journal article by Kalinic and Krisp compares kernel density and Getis-Ord Gi
*
calculations on crime incidents in San Francisco (Kalinic and Krisp 2018). They obtained crime
data from the San Francisco Open Data Portal that included the date of the crime incident, the
category of the crime, the police district in which it occurred, and the latitude and longitude
coordinates of the location. They performed kernel density calculations with several search radii.
Larger search radii produced a smoother density surface, while smaller search radii preserved
more spatial detail. By changing the search radius, they were able to manipulate the size of the
high crime clusters that emerged from the resulting density surface.
While the kernel density is essentially a visualization of the data, a Getis-Ord Gi
*
calculation can increase the understanding of an observed pattern by indicating whether it
exhibits statistically significant clustering or can be explained by a random distribution. To
perform a Getis-Ord Gi
*
calculation on their crime incident data, the authors generated a
hexagonal grid that covered the areas of the city of San Francisco where crimes had been
reported. The grid covered most of the city, but there were no cells in places where no crimes
19
had been reported, such as on freeways and lakes and in parts of parks. They aggregated the
crime data over the grid by calculating the number of crimes reported with coordinates within
each cell. For a grid cell with a large number of crimes to be part of a statistically significant
cluster, it must be surrounded by cells with similar statistics. The results showed two clusters of
grid cells with high crime statistics and less than 1% probability of being due to a random
distribution. These hot spots corresponded to areas of high kernel density, but the authors noted
that not all the areas of high kernel density were statistically significant clusters, and had a higher
probability of being due to a random distribution.
For their Getis-Ord Gi
*
calculation, the authors used the default parameters of their GIS
software. They did not report the size of their grid or the weighting parameters for the
neighborhood used in the calculation. While they explored the effects of changing the parameters
of their kernel density calculation, they did not discuss their Getis-Ord Gi
*
calculation
parameters, which can also affect the results. In their conclusion, the authors suggest that kernel
density and hot spot calculations should be used in conjunction because they can provide
different insights into a spatial distribution.
2.3. Spatial Analysis of Search and Rescue Incidents
Analysis of search and rescue incidents can yield insights into the behavior of lost people,
identify hazardous areas where people become injured and confusing sections of trails where
people lose their way, and can help anticipate where incidents will occur at different times of the
year. The references in this section use GIS to analyze the georeferenced operations of rescue
services.
Continuing the work that was described in section 2.1, Doherty et al. analyzed Yosemite
National Park search and rescue records to understand the applicability of mobility models to the
20
behavior of lost people (Doherty, Guo, Doke, and Ferguson 2014). They considered 213 Parks
Service searches that occurred between 2000 and 2010 for missing people who were involved in
ground-based activities, were last seen in Yosemite National Park, and were found in the park or
within walking distance of it. The places the subjects were last seen and the places they were
found were georeferenced using the point-radius method and used to compare the predictions of
lost person mobility models. A mobility model is an algorithm that attempts to predict the
behavior of a lost person based on their demographic information and the terrain (Koester 2008).
Developing an accurate lost person mobility model is a worthwhile endeavor because search
teams have limited resources, so accurately predicting the most likely route and speed of the
missing person can greatly reduce the area to be searched.
Another application for legacy search and rescue records is to use them to predict where
and when incidents will occur in the future. Doherty et al. used Yosemite National Park search
and rescue records as input for a presence and background learning algorithm, which correlates
the presence of search and rescue incidents with background environmental variables, such as the
elevation, land cover, slope, and distance from trails, roads, streams, and lakes (Doherty, Guo,
Li, and Doke 2014). They georeferenced 2,081 records of search and rescue incidents that
occurred between 2001 and 2010 using the point-radius method, with 95% of the uncertainty
radii falling between 269 and 866 m. The learning algorithm used the points as input, and the
authors state the precision is within acceptable limits based on other published studies.
The authors trained the learning algorithm with the full set of 2,081 records from 2001 to
2010 and calculated a suitability probability, which classified 22.9% of the area of Yosemite
National Park as “suitable habitat” for search and rescue incidents. They compared the locations
21
of 186 search and rescue incidents from 2011 that were not part of the training set, and found
that 91% occurred in areas classified as “likely.”
To see how the spatial distribution of search and rescue incidents changed throughout the
year, the authors separated the rescue records by month and created 12 training sets for the
learning algorithm. The area of the park classified as “likely” varied from 5% in April to 27% in
August. 72% of the search and rescue incidents occurred between May and August, and the
spatial distribution of rescues varied for each month. The learning algorithm classified the
locations of only 64% and 75% of the incidents in January and February as “likely,” but the
result was over 90% for incidents in April, May, June, August, September, November, and
December.
This analysis can help park managers decide where to focus their efforts to prevent
injuries at different times of year and to allocate search and rescue resources in areas where
incidents are likely to occur.
By correlating an incident with background environmental features, the presence and
background algorithm attempts to learn more spatial information from it than its location. The
presence of an incident at one location will increase the suitability of other locations with similar
environmental characteristics. The authors report the sensitivity of their analysis, which is
reflected by the number of incidents whose locations they classify as “likely.” They do not report
any measure of its specificity, which would be reflected by the area classified as likely in which
no incidents occurred.
In addition, and more importantly, the learning algorithm does not consider human
traffic, which is the most important factor leading to search and rescue incidents. The authors
had access to monthly numbers of visitors entering the park, but no finer-grained statistics for
22
traffic at individual features or on individual trails. Yosemite is a vast park, with many remote
areas that require considerable effort to access, and the visitation of different areas varies greatly.
A comparison of the sensitivity of the suitability probabilities calculated by the learning
algorithm and the sensitivity of the suitability probabilities calculated with only the locations of
incidents would provide a measure of how well the environmental variables considered in the
paper represent the combination of environmental hazard and human traffic that leads to search
and rescue incidents.
The United States Coast Guard (USCG) District 14, based in Honolulu, Hawaii, is
responsible for a search and rescue region of more than 12 million square miles of the Pacific
Ocean from near the Philippines to about halfway between Hawaii and the west coast of the
continental United States. A spatiotemporal analysis of historic search and rescue incidents may
help the Coast Guard determine where to position rescue assets at different times of the year. An
article by Hornberger et al. presents a framework for comparing aggregation methods for this
dataset (Hornberger, Cox, and Hill 2019).
The authors compare quadrat methods, in which square grids are laid over the entire
study area, with zonal distribution models, which can have zones of different areas depending on
factors such as the distance from land or historical trends in distress calls. They use a training
dataset of 2629 search and rescue missions that occurred from 2011-2015 and compare the
predictions of their model to a test dataset of 1080 incidents that occurred from 2016-2017.
The authors generated six quadrat grids, five with uniform cells named grids A through E,
and one named grid F in which the cells containing Guam and the Hawaiian islands, where the
greatest number of incidents occur, were further divided into 1
0
squares. The quadrat grids had
between one and 212 zones. They also generated two zonal distributions based on k-means
23
clustering algorithms, one based on the number of resources assigned to historical rescues and
one based on the ranges of Coast Guard rescue assets, such as ships, helicopters, and airplanes.
The authors measured the accuracy of the aggregation methods by calculating distance-
and volume-based errors. The distance-based aggregation error is the sum of the differences
between where incidents were modeled as occurring and where they actually occurred, and the
weighted version is weighted by the number of assets deployed to each incident. The authors
found that the distance-based aggregation error decreased with increasing number of zones for
the quadrat methods, and that the k-means clustering methods had smaller errors than the quadrat
grids with similar number of zones.
The volume-based aggregation error is the difference between the monthly demand
predicted by a deterministic model based on the training set of incidents and the actual demand
calculated from the test set. The authors found the volume error increased with the number of
zones for the quadrat grids and that the k-means clustering grid had a greater volume error than
the quadrat grids with similar numbers of zones, perhaps because the greater number of zones
gave the algorithm more opportunities to over- or under-predict the actual demand.
The references in this chapter that explored the MAUP all used the aggregation with the
smallest zones, either census blocks, TAZs, or the incident points themselves, as the “gold
standard” against which they compared other aggregation methods. They examined ways in
which larger and fewer zones could be used while preserving the results of statistical calculations
or minimizing the error introduced by changing the representation of locations. If the memory
and computer time required for the finer aggregation are reasonable, however, there is no reason
to compromise the accuracy of the results by using a more coarse aggregation.
24
Chapter 3 Methodology
This chapter reviews the methodology for the project. Section 3.1 discusses the construction of
the rescue geodatabase, section 3.2 discusses the aggregation of the rescue statistics over a grid
and the statistical calculations carried out using it, and section 3.3 discusses the kernel density
calculations carried out to visualize the distribution of rescues.
3.1. Rescue Geodatabase
The goal of the database was to support the analysis of the hoist rescue records to find
spatial and seasonal patterns to, in turn, improve readiness and recommend preventative
measures. The attributes of the rescue records necessary for this analysis were the date, number
of victims rescued, location, and elevation. The accuracy and precision of the locations
associated with the rescue records is of utmost importance to support spatial analysis. Making
use of the other attributes recorded in the rescue documentation, such as the aircraft, personnel,
and rescue devices used, is beyond the scope of this project.
3.1.1. Sheriff’s Aviation Documentation
The dataset that inspired this project is the sheriff’s aviation hoist log books. Each hoist,
which is a piece of equipment that can be transferred between helicopters, has a paper log book,
in which an entry is recorded every time it is used. The books are used to track issues and
damage and schedule maintenance tasks such as inspecting and replacing cables. The most recent
log books record the date; the purpose of the hoist; the number and running total of evolutions;
the length of cable used; the aircraft and the number of hours on it; the personnel; the location
and elevation; and a narrative. Figure 5 shows a blank page of a hoist log book.
25
Figure 5 A page of a hoist log book.
Recently, members of the air rescue team obtained access to the archive of hoist logs,
which consisted of 33 books. They transcribed the mission hoists into an Excel spreadsheet,
which became the basis for this project. The spreadsheet has the attributes listed above and some
others that were found in the narrative sections. The additional attributes in the Excel spreadsheet
include whether the hoist occurred during the day or night, the outside air temperature, the
26
number of victims and dogs rescued, the rescue devices used, whether the rescue was a pickoff,
whether the rescue was assisted by a ground team, whether hover steps were performed, whether
a static discharge cable was used, and whether the hoist was an insertion of the sheriff’s dive
team. The records in the log books are of varying completeness, with some providing
information on every attribute listed above, while others only have a date and list of personnel.
There are 921 records in the spreadsheet. Figure 6 shows the format of the Excel spreadsheet.
Figure 6 Format of the Excel hoist rescue spreadsheet.
The fields of the spreadsheet that pertain to this project are Date, Location, Elev, and
Vic’s. The Date field records the date of the rescue in MM/DD/YYYY format, which is
recognized by the Excel and ArcGIS Pro programs, so the records can be sorted and selected by
date. The first rescue was recorded on December 1, 1993 and the most recent rescue included in
this project was recorded on September 5, 2020. The Vic’s field records the number of victims
rescued. Seven hundred sixty-nine records have one victim. The greatest number of victims
recorded for a rescue is eight.
The Elev field records the elevation of the rescue in feet. The lowest elevation recorded is
500 ft. and the highest is 10,700 ft. The precision of the elevation data is not immediately
apparent, as there are no written guidelines in the aviation unit as to the proper source of
elevation data or how it should be recorded. One likely source of elevation data for recent hoists
is Google Earth, as the program is currently installed on the computers at the aviation unit and it
is standard procedure to check the elevation at the coordinates when they are supplied with the
27
dispatch information. The aviation unit was performing hoist rescues before Google Earth was
available, however, so it cannot be assumed to be the source for all elevation data.
Another likely source of elevation data is the pilot, as it is an important factor in their
consideration of the capability of the helicopter to perform the hoist, so they are very aware of
their altitude and how much power the helicopter requires to achieve a stable hover and complete
the rescue.
The level of rounding of the elevation is at the discretion of the crew member completing
the hoist record, usually the crew chief. Of the 921 rescue records, 80 have no elevation
recorded, three have elevations recorded to the ones place, 11 have elevations recorded to the
tens place, 543 have elevations recorded to the hundreds place, 280 have elevations recorded to
the thousands place, and four report elevations of “10,000 ft.” Because the majority of the
elevations are reported to the hundreds place, an error of
+
/- 100 ft. was assumed as the precision
of the elevation data.
The Location field of the spreadsheet contains a text description of where the rescue
occurred. None of the hoist records include coordinates. Like the elevation, the location
description is recorded at the discretion of the crew member completing the hoist record. There
are no set standards in the sheriff’s aviation unit as to the nomenclature or extent of the described
locations. The descriptions of the locations depend on the training lineage of the crew member,
the colloquial location names they prefer, and the level of precision with which they choose to
describe where the rescue took place. Thirty-nine rescues have no location recorded, and several
are described with very general locations. For example, “Joshua Tree National Park,” “Riverside
County,” and “San Gorgonio Wilderness” are the recorded locations for rescues in the
spreadsheet. Locations such as these are too general and imprecise to support the spatial analysis
28
of this project, so records had to be discarded from the database if information to improve the
precision of their locations could not be found.
In addition to the hoist log books, an additional form of documentation called an
“Unusual Occurrence Report,” commonly referred to as a “blue card,” is recorded for some
rescues. The purpose of the blue cards is to account for aircraft and personnel time when they are
spent on missions outside the usual scope of the aviation unit. If the victim of the rescue is not a
San Bernardino County resident, section 26614.5 of the California Government Code states
“[t]he county or city and county of residence of a person searched for or rescued by the sheriff
under the authority of Section 26614 shall pay to the county or city and county conducting such
search or rescue…all of the reasonable expenses…of such search or rescue.” (California
Legislative Information 2020) In this case, the blue card records the aircraft and personnel hours
for the purpose of billing the victim’s county of residence for the rescue. The format of the blue
card has changed several times throughout the study time period, with the most recent version
recording the date, GPS coordinates, personnel, aircraft, victim information, and a narrative
describing the incident. Figure 7 shows the most recent blue card format.
An archive of 520 blue cards was obtained from sheriff’s aviation. Many of the blue
cards describe activities other than hoists, such as hover steps, rescues in which the helicopter
was able to land, inter-facility transports, and time spent assisting searches. Some of the blue
cards that document hoist rescues, however, contain GPS coordinates or descriptions of the
rescue in the narrative section that improve the precision of the location in the record.
The sheriff’s aviation documentation described in this section are a maintenance record
and an accounting form. They contain geographic information, but were not collected with the
intent of making maps or being the basis for spatial analysis. Although the records were recorded
29
in a somewhat haphazard fashion and without geographic best practices in mind, the subsequent
sections demonstrate how different attributes and data sources can be combined to produce
spatial information of sufficient precision to support analysis that can identify spatial and
temporal trends, improve preparation, and direct preventative measures.
Figure 7 Most recent Blue Card format.
3.1.2. Database Construction
The first step in constructing the rescue database was to georeference the locations
described in the hoist records. To create the most precise representation of the described
locations, the shape method was used, in which a polygon represents the location of each rescue,
as described in section 2.1. Two hundred eighty-seven unique location descriptions were
identified from the 921 rescue records. Topographic maps were downloaded from the USGS
National Map Viewer (USGS 2020) and printed. Experienced members of the aviation unit
30
assisted in finding the described locations on the topographic maps. One member, in particular,
has been a volunteer crew chief and air medic since the inception of the program and has
participated in hundreds of rescues. Prior to joining the air rescue team, he was a member of a
San Bernardino sheriff’s ground search and rescue team and a patrolman with the forest service.
He was very familiar with the locations described in the hoist records and spent hours identifying
them on the topo maps. The locations were confirmed with internet searches, with many trip
reports and blog posts providing clues as to the vernacular place names used by different groups.
For example, the dirt bike community refers to a section of Forest Service Trail 1W17 as the
“Malcolm Smith Trail,” while the topographic maps label it the “Redonda Ridge Trail.” All three
names appear in the hoist rescue records but could be consolidated to one feature after
confirming they referred to the same feature.
Once they had been identified on the paper topo maps and confirmed with internet
searches, the list of unique locations was imported into ArcGIS Pro and used to create a feature
class. A polygon was created for each location, with the goal of creating the smallest polygon
that covered everywhere a crewmember recording a rescue in a log book would associate with its
name. Some degree of subjectivity was involved in creating the polygons. The boundaries for
canyons were defined by following the ridge lines on either side. The boundaries for linear
features such as trails, roads, and creeks were defined by creating a 50 m buffer on each side for
their entire lengths. The boundary lines for peaks were drawn by identifying the saddle points
that separated them from their neighboring peaks and roughly following contour lines to connect
them.
After polygons were created for the unique locations, the rescue spreadsheet was joined
to the locations feature class with a one-to-many join, resulting in a new rescue feature class
31
where each rescue record had a polygon geometry corresponding to its described location.
Thirty-nine rescues had no location recorded and were discarded.
The next step was to parse the blue cards for those describing hoist rescues and examine
them for useful information. Thirty-four blue cards were found to correspond to hoist rescues
and have information not recorded in the log book. Nine blue cards reported latitude/longitude
coordinates, four in degrees minutes seconds format, one in decimal minutes, and four in decimal
degrees. Since WGS84 is the default datum for Google Earth, Google Maps, and iPhone and
Android GPS systems, it was assumed to be the datum of the blue card coordinates. The
coordinates were plotted and found to be consistent with the blue card narratives and locations
and elevations contained in the hoist records, so the polygons for those rescues were changed to
small circles enclosing the coordinate points. Three blue cards included elevations that were not
reported in the hoist records, so they were filled in. The 25 blue cards that did not have
coordinates had descriptions in their narrative sections that allowed the rescue locations to be
determined more precisely.
For example, the narrative section of one blue card contained the text, “dispatched to
Cucamonga Canyon for person that had fallen off cliff laying on edge of water.” The hoist record
spreadsheet listed a rescue occurring on that date at 3000 ft. in Cucamonga Canyon. The
combination of the location description, the elevation, and the narrative stating the victim was
next to the water allowed the polygon for the rescue to be modified to a very small area. Figure 8
shows the process of georeferencing this rescue with Cucamonga Canyon in blue, the area
between 2900 and 3100 ft. in the canyon in red, and the part of the red area near the creek in the
bottom of the canyon in yellow.
32
Figure 8 Georeferencing process for a rescue in Cucamonga Canyon.
33
Another blue card narrative contained the text, “Victims located by 40K5 two peaks east
of Cucamonga Peak at the 8500' level.” The hoist record for that date indicated a rescue on the
Cucamonga Peak Trail. It was a simple matter to follow the trail, which was on the topographic
map, east from Cucamonga Peak to the second peak and find the closest intersection of the trail
with the 8500 ft. contour line.
When the rescue locations had been modified to include all of the blue card information,
the next step was to modify them using the elevation field of the hoist records. The intersection
of the described location and reported elevation was found for each record. Many of the
described locations were linear features such as hiking trails, creeks, ridges, or sections of road
that start at a low elevation and climb continuously to a peak or junction. In these cases, the
described location intersects the reported elevation at one point and the location polygon can be
modified. The uncertainty in the elevation data was assumed to be +/-100 ft., and an
approximately 50 m buffer was drawn around the linear objects, which resulted in very precise
location polygons for the rescue records.
After the information in the hoist records and blue cards had been included in the
database, sheriff’s aviation personnel who were listed as participating in rescues with
documentation that was incomplete or had imprecise location descriptions were consulted to see
if they could provide additional information as to where the rescues took place.
One retired pilot, in particular, had extensive personal records that made it possible to
complete several hoist rescue records with missing information. He used Google Earth to point
out the locations of rescues whose described locations were not precise or had not been found
with web searches and consultations with other team members.
34
During the process of individually examining the rescue records, those with location
descriptions that were too vague to be definitively determined after the blue cards, elevation
information, and sheriff’s aviation personnel were considered were discarded. For example, three
rescue records had “29 Palms USMC Base,” listed as their location, 14 had “Joshua Tree
National Park,” and 10 had “San Gorgonio Wilderness.” During this step, 139 records were
eliminated, leaving 743 rescue records in the database. The records that remain in the database
are an incomplete representation of the operations of the aviation unit, and high density clusters
may be missing from the analysis, but without more precise location information, there is no way
to include them. If geographic coordinates are collected for future rescues, the georeferencing
process can be eliminated, the precision of the locations will be improved, and all records will be
included.
At this point, the 95
th
percentile of the areas of the polygons representing the locations of
the rescues was approximately 48,000 square meters, which corresponds to a square with 220 m
sides or a circle with a radius of 124 m. The process of creating the polygons for the rescue
records was not entirely objective. Some interpretation was involved in determining the extent of
the described locations and the meaning of the blue card narratives. The combination of the
described locations, elevations, blue card information, and consultation with sheriff’s aviation
personnel, however, allowed the locations of 743, or 81%, of the 921 rescue records to be
precisely determined.
3.1.3. Selecting Hiking and Climbing Records
The intent of this project is to focus on rescues of hikers and climbers. To eliminate
rescues involving cars, off road vehicles, and people in populated areas, such as flood control
channels, a TIGER/Line shapefile of roads in San Bernardino County was downloaded from the
35
U.S. Census Bureau’s Web Interface (US Census Bureau 2019). Features classified as primary,
secondary, local, and private roads; 4wd trails; ramps, service drives, and parking lots; and
stairways and alleys were selected and used to create a new feature class of roads. Features
classified as walkways, pedestrian trails, bike paths, and bridle paths were excluded. Some
hiking trails that were incorrectly classified as roads were deleted from the feature class. Rescues
located within 50 m of features in the roads feature class were selected and eliminated. In
addition, rescues whose locations were identified by road mile markers, which are almost always
cars over the side, were eliminated. Eighty records were eliminated from the database in this
step, leaving 663.
The rescue database, with location polygons providing a representation of where the
rescues happened, is the foundation for the analysis phase of the project. Understanding the
process of determining the locations of the rescues will inform the decisions made about what
types of analysis can be carried out to get the greatest benefit from this information.
3.2. Spatial Analysis of Rescues
This section describes the spatial analysis of the rescue records. Section 3.2.1 discusses
the aggregation of the rescue statistics over a hexagonal grid and visualization and analysis of
rescues by elevation. Section 3.2.2 presents the Moran’s I calculation to determine the spatial
autocorrelation of the rescues, and Section 3.2.3 discusses the hot spot calculations, which use
the Getis-Ord Gi
*
statistic to identify significant clusters of values. Section 3.2.4 concludes with
the subsequent temporal hot spot calculations and the statistics gathered on the areas of highest
concentration of rescues.
36
3.2.1. Hexagonal Grid and Visualization of Rescues by Elevation
To provide a framework for the visualization and analysis of the rescue data, a hexagonal
grid with cells of 50,000 m
2
area was generated over the area in San Bernardino, Los Angeles,
and Riverside counties bounded by the northernmost, southernmost, easternmost, and
westernmost rescues in the geodatabase. The cells have sides approximately 140 m long. The cell
size was chosen to be slightly larger than the 95
th
percentile area of the rescue location polygons,
so the rescues can be aggregated into the appropriate cells while preserving as much spatial
detail as possible. The area of the grid cells is larger than the uncertainty in the location of 95%
of the rescues, it is the finest possible grid that can be supported by the precision of the data.
Initially, 1,215,869 cells formed the grid.
In order to support accurate spatial statistical calculations, the extent of the grid must
reflect the areas where hoist rescues of hikers and climbers can happen. The purpose of the
calculations is to analyze the distribution of where rescues did happen within the area where they
can happen. As was discussed in section 3.1.3, rescues occurring in populated areas and near
roads were not considered. To remove these areas from the grid, cells within 50 m of the roads
feature class used to select and eliminate rescues were deleted.
Another area where hoist rescues cannot happen is where it is flat enough for a helicopter
to land. To identify and remove areas of flat terrain from the grid, a 1/3 arc-second digital
elevation model (DEM) was downloaded from the USGS National Map (USGS 2020). This
DEM has a resolution of approximately 10 m. A slope was calculated from the DEM and areas
with slope less than six degrees were eliminated from the grid. There are many factors that
influence a helicopter’s ability to land, such as the wind, elevation, visibility, and ground cover.
The sheriff’s department helicopters should consistently be able to land in areas with slope of
37
less than six degrees. In addition to flat areas, this step eliminated bodies of water from the grid,
which are also areas where hoist rescues cannot happen.
After these steps, 357,805 grid cells remained. These cells cover the parts of the study
area more than 50 m from roads and with slope 6 degrees or greater. The extent of the grid is
shown in blue in figure 9.
The grid had the smallest cell size that could be supported by the precision of the
locations of the rescues, with homogeneous cells throughout its extent. The grid is the closest
possible representation of the pattern of the rescue locations and preserved as much spatial detail
as possible. The memory and computer time required for the spatial statistical calculations were
reasonable, so there was no need to use a coarser grid, which would increase the error in the
representation of the rescue locations and possibly change the results, as discussed in chapter 2.
The rescues were aggregated over the final grid by joining the rescues to the cells within
which the centroid of their polygons were located. The total number of victims for the rescues
joined to each cell was calculated. The greatest number of victims for one cell was 54. Three
hundred cells had one or more victims, and the rest had zero. Figure 9 shows the hexagonal grid
and the rescue location polygons in six areas of high concentration of rescues that are
investigated in detail in the upcoming chapters. The rescue location polygons are gold and
become darker when they overlap.
As requested by the leadership of the aviation division, the rescues above and below
6,000 ft. were visualized so their numbers and locations could be compared. To accomplish this
visualization, the grid was symbolized with circles proportional to the number of victims for the
rescues in each cell, with different colors for rescues above and below 6,000 ft.
38
Figure 9 Rescue location polygons and hexagonal grid.
39
To demonstrate the trends in elevation of rescues, charts were generated for the numbers of
rescues above and below 6,000 ft. for the years since the inception of the program, the months of
the year, and the different types of helicopters.
3.2.2. Moran’s I Calculation
The global Moran’s I statistic measures whether a spatial distribution is clustered,
random, or dispersed. It is defined by the following formula:
𝐼 =
∑ , ( ̅) ̅ ∑ ( ̅)
,
(2)
where 𝑛 is the number of features, 𝑥 is the value of variable 𝑥 for the feature 𝑖 , 𝑤 is the
statistical weight between features 𝑖 and 𝑗 , 𝑊 is the sum of all the 𝑤 , and 𝑥 ̅ is the mean value of
the variable 𝑥 (Anselin 1995).
The 𝑍 score for the statistic is calculated by the formula
𝑍 =
( )
( )
,
(3)
where
𝐸 (𝐼 ) =
−1
(𝑛 − 1)
(4)
and
𝑉 (𝐼 ) = 𝐸 (𝐼 ) − 𝐸 (𝐼 )
.
(5)
The global Moran’s I formula calculates the cross product between the deviation of the
value of a point from the mean and the deviations of the values of the points near it. The values
that are calculated and summed are determined by the matrix of statistical weights. For example,
if the user wants features within a distance d of each other to be considered equally in the
40
Moran’s I calculation, the 𝑤 will be one if points i and j are within d of each other and zero
otherwise.
If a point with a value higher than the mean has neighbors whose values are also higher
than the mean, their cross products will be positive. Similarly, if a point with a value lower than
the mean has neighbors with values lower than the mean, their cross products will also be
positive. If a point with a value higher than the mean has neighbors whose values are lower than
the mean, their cross products will be negative. The cross products of a point with neighbors
whose values are both higher and lower than the mean will cancel each other out and not
contribute to the overall sum. As a result, a positive value of the I statistic indicates that points
with similar values are clustered together, a value near zero indicates the values are randomly
distributed, and a negative value indicates the values are dispersed, meaning that points tend to
be surrounded by points with dissimilar values.
To determine the statistical significance of the I statistic, the distribution is compared to
the null hypothesis, which is that the distribution is the result of a random process with no pattern
or correlation. For a set of n features, there are n! possible permutations of the values of the
features. The expected value and variance of the I statistic for these permutations are calculated
and compared with the I statistic of the actual distribution. The 𝑧 score is the number of standard
deviations the I statistic of the distribution is from the expected value. The p-value indicates the
probability that the distribution is random. A low p-value means that there is a low probability
the distribution is random and the null hypothesis can be rejected.
To characterize the degree to which the distribution of rescues is clustered, random, or
disperse, a Moran’s I calculation was carried out using the hexagonal grid described in section
3.2.1. The calculation used a second-order adjacency network with a fixed distance band, so for a
41
target cell, its first and second nearest neighbor cells were given a statistical weight of one and
all other cells were given a weight of zero. Figure 9 shows the second-order adjacency network
for a target cell near Mt. San Antonio. The network size was chosen to be of the approximate
scale of the canyons and other features in the study area, so that the rescues described as
happening in one location could become a cluster, but rescues in neighboring locations would
not contribute. As can be seen in Figure 9, the areas of Mt. San Antonio and Icehouse Canyon
are greater than one network, while the areas of the other locations are smaller.
3.2.3. Getis-Ord Gi
*
Calculation
The Getis-Ord Gi
*
statistic identifies statistically significant hot and cold spots by
calculating the degree of clustering around every feature (Getis and Ord 1992). While the
Moran’s I statistic is a global statistic, which returns one set of results for the entire pattern, the
Getis-Ord Gi
*
statistic is calculated for each feature individually. The resulting pattern shows
where statistically significant clusters of high values, or hot spots, and low values, or cold spots,
are located.
The Getis-Ord Gi
*
statistic is calculated by the following formula:
𝐺 ∗
=
∑ 𝑤 𝑥 − 𝑥 ̅ ∑ 𝑤 𝑆 𝑛 ∑ 𝑤 − ∑ 𝑤 𝑛 − 1
,
(6)
where 𝑛 is the number of features, 𝑥 is the value of variable 𝑥 for the feature 𝑖 , 𝑤 is the
statistical weight between features 𝑖 and 𝑗 , and 𝑥 ̅ is the mean value of the variable 𝑥 , and
42
𝑆 =
∑ 𝑥 𝑛 − 𝑥 ̅ .
(7)
The Gi
*
statistic is a z-score. Essentially, it indicates how many standard deviations the
proportion of the total value of the variable x that is found near feature i is from the expected
value.
To identify statistically significant hot spots, a Getis-Ord Gi
*
calculation was carried out
using the hexagonal grid. As with the Moran’s I calculation, a second-order adjacency network
was used with a fixed distance band. The locations with the highest Gi
*
values were identified for
further study.
3.2.4. Temporal Density and Hot Spot Calculations
The areas with the six highest Gi
*
values from the hot spot calculation were identified to
analyze for seasonal trends. The rescues in the database were separated by months, and a hot spot
calculation with the same parameters described in the previous section was carried out for each
month. Statistics were gathered for the rescues occurring in the top six areas for each month to
accompany the hot spot results. The resulting seasonal trends can help anticipate operations and
guide preventative efforts to the areas and times where the greatest number of rescues occur.
3.3. Kernel Density
A kernel density calculation places a circular kernel function on each point in a
distribution. The kernel function has a smoothly curved surface, has its maximum value at its
origin, and goes to zero at a defined search radius. If a weighting attribute is specified, the area
under the function centered on a point is equal to the attribute value for that point. If no attribute
is specified, the area under the function is equal to one. The kernel function implemented in
43
ArcGIS Pro is a quartic function (Silverman 1986). Figure 10 shows a one-dimensional quartic
kernel function with a search radius of one and an area of one (Amberg 2021). The value of the
density at a point (x,y) is the sum of the values of the kernel functions centered on points within
the search radius:
𝐷 (𝑥 , 𝑦 ) =
∑ 𝑥 1 − , for 𝑟 < 𝑟 ,
(8)
Where 𝐷 is the kernel density, 𝑖 = 1, … , 𝑛 are the input points, 𝑟 is the search radius, 𝑥 is the
value of the attribute x of point i, and 𝑟 is the distance between (x,y) and point i.
Figure 10 quartic kernel function
To represent the density, a raster, or grid, is defined over the study area. The value for
each raster cell is calculated by evaluating the density at the center of the cell. Kernel densities
are used to create a generalized view of a set of point data. Larger values of the search radius
44
spread each point out over a greater area and give a smoother density surface, while smaller
values retain more detail. A smaller raster cell size will give a finer-grained view of the density
surface at the expense of using more computing time and storage space.
For this project, kernel densities were calculated in conjunction with the hot spot
calculations to provide two different perspectives on the distribution of rescues. A point was
placed at the centroid of the location polygon for each rescue record. Using these points, a kernel
density was calculated with a search radius of 2200 m and the number of victims as the specified
attribute. The density was represented with a square grid with 500 m sides. The search radius
was chosen to create a density general enough to allow rescues near each other to combine while
preserving the spatial pattern. The precision of the location information is more than sufficient to
support these parameters. The resulting raster shows a density detailed enough to identify areas
of high concentration of rescues and contrast the density with the results of the hot spot
calculation.
To accompany the temporal hot spot calculations, kernel density calculations were also
carried out for each month using the same parameters as for the overall dataset.
45
Chapter 4 Results
This chapter describes the results of analysis of the rescue geodatabase. Section 4.1 discusses the
trends in the elevations of the hoist rescues in the database. Section 4.2 presents the results of the
hot spot and kernel density calculations, and section 4.3 presents the results of the subsequent
temporal calculations carried out on the areas with the highest Gi
*
scores.
4.1. Trends in the Elevation of Hoist Rescues
Figure 11 shows the hoist rescues in the database, aggregated over a hexagonal grid and
symbolized with circles proportional to the total number of victims in each cell as described in
section 3.2.1. The vast majority of hoist rescues occur in the mountains in the southwest part of
the county. Rescues below 6,000 ft. are blue and rescues above 6,000 ft. are orange. The main
areas where rescues above 6,000 ft. have occurred are near Mt. San Antonio, San Gorgonio
Mountain, and Big Bear, with a few near Mt. San Jacinto in Riverside County and one on Clark
Mountain near the Nevada border. The hoist rescues below 6,000 ft. are concentrated in the
lower areas of the mountains. Lytle Creek and Deep Creek are two areas that account for many
of the lower elevation rescues. Figure 12 shows a series of charts with statistics for rescues above
and below 6,000 ft. In every chart, rescues below 6,000 ft. are blue and rescues above 6,000 ft.
are orange.
The top left chart in Figure 12 shows the numbers of rescues above and below 6,000 ft.
for each year from 1994 to 2019. From 1994 to 2008 the aviation unit performed ten to fifteen
hoist rescues below 6,000 ft. and five to ten hoist rescues above 6,000 ft. per year. In 2008 the
aviation unit began to revise its hoist rescue procedures and pursue them more aggressively,
which led to an increase in the number of hoist rescues below 6,000 ft. from nine in 2008 to 37 in
46
Figure 11 Hoist Rescues above and below 6,000 ft.
47
2012. From 2010 to 2019 the aviation unit performed between 26 and 46 hoist rescues below
6,000 ft. per year. After an approximately five-year lag, the number of rescues above 6,000 ft.
also increased, from seven in 2010 to 44 in 2017. From 2015 to 2019 the aviation unit performed
between 34 and 44 hoist rescues above 6,000 ft. per year.
The bottom left chart of Figure 12 shows the numbers of hoist rescues by month. July has
the greatest number of hoist rescues below 6,000 ft., with the numbers generally increasing
during the spring, at their highest levels in the summer, decreasing in the fall, and at their lowest
levels in the winter. The hoist rescues above 6,000 ft. follow a similar trend, except that January
has the fourth highest monthly total. The highest monthly totals are in June, July, and August,
when the higher temperatures will have the greatest effect on the density altitude as described in
Section 1.5.
Figure 12 Statistics of rescues above and below 6,000 ft.
48
The charts on the right side of Figure 12 show the proportion of rescues performed by the
medium helicopters and patrol helicopters. The medium helicopters have performed 761 hoist
rescues, with 43% of them above 6,000 ft. The patrol helicopters have performed 118 hoist
rescues, with 32% of them over 6,000 ft.
The results of this section clearly demonstrate a significant and sustained need for hoist
rescues above 6,000 ft., with the greatest demand during the hottest part of the year. A rescue
helicopter that can only perform hoist rescues below 6,000 ft. will be limited in its ability to
provide this service in the county.
4.2. Statistical and Kernel Density Calculations
To determine the spatial autocorrelation of the distribution of rescues, the rescues were
aggregated over a hexagonal grid with 50,000 m
2
cells as described in section 3.2.1. A global
Moran’s I calculation was carried out using a second-order adjacency network and a fixed
distance band. The resulting 𝑧 was 45.0 and the p-value was 0.00, meaning the calculated I
statistic was 45 standard deviations from the expected value for a random distribution, which
indicates statistically significant spatial autocorrelation.
To identify statistically significant rescue hot spots, a Getis-Ord Gi
*
calculation was
carried out using the same hexagonal grid, second-order adjacency network, and fixed distance
band. The results are shown in the bottom inset of figure 13. Cells with Gi
*
scores greater than 10
are shown in red, with the color becoming darker with greater Gi
*
score. Because their Gi
*
scores
are so high, the p-value for all of these cells is 0.00. The hexagons shown are larger than those of
the grid because the actual grid cells are too small to be seen at the scale of the inset map.
As described in section 3.2.2, a point was placed at the centroid of the location polygon
of each of the 663 rescue records in the geodatabase. A kernel density was generated with the
49
number of victims as the weighting attribute, a 2200 m search radius, and a 500 m cell size.
Because the number of victims was used as the weighting attribute, a rescue in which two
victims were rescued contributes twice as much to the density as a rescue with one victim.
After it was calculated, the density raster was symbolized with a color gradient, with low
densities being transparent, then becoming blue, red, and yellow with increasing density. The
units of the kernel density are victims rescued over the study time period per square meter. This
is not a meaningful objective quantity, but the kernel density provides a qualitative indication of
areas with a high concentration of rescues. The kernel density is shown in the top inset of Figure
13. The six areas with the highest Gi
*
scores are labeled in both insets. The next few paragraphs
introduce these areas from west to east:
Mt. San Antonio, also known as Mt. Baldy, is a very popular mountain for hiking,
climbing, and skiing. The highest Gi
*
score on Mt. Baldy is 67.7. It is a very prominent peak,
with summit elevation 10,069 ft. It is visible from a large portion of southern California and is
usually covered with snow during the winter. The closest ski area to Los Angeles is located on
the east side of the mountain, and a large bowl on the south face is used for backcountry skiing.
The Sierra Club maintains a ski hut in the bowl at an elevation of 8,300 ft. A section of trail
called the Devil’s Backbone traverses a ridge between approximately 8,600 and 9,200 ft. with
steep drop-offs on both sides that connects the ski area to the summit. The steep and icy terrain
in these areas during the winter results in falls that can end in serious injuries and fatalities.
During the summer, hikers often lose the trail during their descent from the summit, go down the
wrong canyon, and find themselves unable to continue when it gets dark. The combination of an
attractive and highly visible mountain, hazardous terrain, and easy access to a large population
50
Figure 13 Rescue kernel density and Gi
*
scores.
51
results in many search and rescue operations throughout the year, some of which become hoist
rescues.
Icehouse Canyon is located along the road that leads to the Mt. Baldy trailheads and ski
area. It branches off San Antonio Canyon at approximately 5,000 ft. elevation and climbs to a
ridge, intersecting it at Icehouse Saddle at approximately 7,600 ft. The ridge connects to the ski
area and the summit of Mt. San Antonio. There are several north-facing side canyons that are
popular for ice climbing in the winter. The highest Gi
*
score in Icehouse Canyon is 35.1.
Bonita Falls is a series of waterfalls located on the south fork of Lytle Creek. The lowest
waterfall drops approximately 160 ft. to the bottom of Bonita Canyon, which is at 3,100 ft.
elevation and is accessible by a
3
/4 mile hike from the road. There are several cascades higher in
the canyon, with a total height of almost 500 ft. Bonita Falls is heavily impacted by human
activity, with abundant graffiti and trash during the summer, when there is the most traffic. It is
possible to climb past the lowest waterfall to see the other waterfalls in higher parts of the
canyon, which are up to approximately 3,600 ft. elevation, but the trails are extremely steep and
narrow. Hikers often fall and get injured or become stuck when they are unable to continue
climbing up or climb back down. The highest Gi
*
score at Bonita Falls is 79.8.
The Deep Creek Hot Springs are geothermal hot springs located along Deep Creek.
Several pools have been constructed in which the temperature is controlled by mixing the hot
spring and creek waters. The shortest hike is from the north, with a 1.8 mile trail that descends
almost 1000 ft. from a ranch outside of Hesperia to the hot springs at 3600 ft. The trail is steep,
sandy, and strenuous, with no shade in the summer. The Pacific Crest Trail also passes by the hot
springs. The hot springs are clothing optional, and there are several cliffs that people jump off
into the creek. The combination of natural beauty, party atmosphere, alcohol and drug use, and
52
hazardous terrain make this a frequent location for hoist rescues. The highest Gi
*
score at Deep
Creek is 31.2.
Aztec Falls is a popular swimming hole along Deep Creek near the town of Lake
Arrowhead at approximately 4,500 ft. elevation. There are a number of cliffs that people jump
off into the water, which can lead to injuries. It is accessible by a ½ mile hike along the Pacific
Crest Trail from a parking area. On May 22, 2020, the Forest Service closed Aztec Falls due to
human impact, including illegal campfires, graffiti, trash, and gridlock in the parking area and
surrounding Forest Service roads that has prevented ambulances from responding to calls (USFS
2020). Between the closure date and September 2020, two hoist rescues were recorded at Aztec
Falls, compared to four in 2019, four in 2018, twelve in 2017, and six in 2016. The highest Gi
*
score at Aztec Falls is 63.8.
Big Falls is the tallest permanent waterfall in southern California. It is located on the edge
of the San Gorgonio Wilderness near the town of Forest Falls. The lowest waterfall has a drop of
approximately 200 ft. The observation area for the waterfall is at approximately 6,000 ft. and is
reached by a ½ mile hike from a parking lot. Like Bonita Falls, it is possible to climb past the
first waterfall to the higher waterfalls in the canyon, which go up to approximately 6,400 ft. The
trail is narrow, steep, loose, and makes several stream crossings. Due to the number of rescues in
the upper part of the canyon, the Forest Service closed Big Falls beyond the observation area in
July 2019 (Gundran 2019). The closure was renewed in May 2020 until May 2021 (USFS 2020).
Since the closure, there have been three hoist rescues in 2019 and three in 2020, compared with
ten in 2018, none in 2017, and eight in 2016. Big Falls has the highest Gi
*
score in the study area
at 97.1.
53
The six locations identified in this section make up a tiny fraction of the area of the
county, but account for 365 of the 937, or 39%, of the victims rescued via hoist during the study
time period, according to the database. By focusing on these six areas, the team can train and
prepare for more than a third of its rescues. These areas may also provide opportunities to
mitigate hazards to prevent injuries. In the next section, hot spots, densities, and statistics for the
hoist rescues in these six areas are presented for each month to identify seasonal trends.
4.3. Monthly Hot Spots and Kernel Densities of Rescue Incidents
The database records were separated by month and hot spot and kernel density
calculations were carried out on a study area containing the top six locations. The hot spot
calculations used the same grid, second-order adjacency network, and fixed distance band as the
calculations on the overall data set. The kernel densities used a search radius of 2200 m and a
cell size of 500 m, the same parameters as the analysis of the whole dataset. The symbology of
the hot spots and densities were adjusted from the previous section because the Gi
*
scores and
densities were different after being separated into months, and are the same for all twelve
months. The grid cells were once again symbolized with hexagons larger than the actual grid
cells so they would be visible in the inset maps. A chart with the number of victims rescued in
the top six locations was generated for each month and included with the figures. Locations with
more than five rescues for the month were labeled on the hot spot and density maps.
Figure 15 shows the hot spots and kernel densities for March, April, and May, which are
relatively slow months for hoist rescues. The kernel densities are in the left insets and the hot
spots are on the right. In March, the greatest number of rescues were on Mt. San Antonio and
54
Figure 14 Kernel densities and hot spots for March, April, and May.
55
Bonita Falls. The Gi
*
score on Mt. San Antonio is 52.6 and the Gi
*
score for Bonita falls is 122.
The high number of rescues at Bonita Falls is somewhat unexpected because, as a waterfall and
swimming destination, it should see its highest traffic and number of rescues in the summer. The
high number of rescues in March can be traced to an incident on March 20, 2016, when five
victims were rescued, and another on March 20, 2019, when three victims were rescued. These
two incidents account for eight of the eleven rescues at Bonita Falls in March.
In April, the highest Gi
*
score is at the Deep Creek spillway, which is not one of the top
six locations identified by the overall hot spot calculation. The Gi
*
at the Deep Creek spillway is
105. Four hoist rescues have been recorded at that location in April, with five victims being
rescued on April 16, 2020. The other three rescues had one victim each.
In May, the greatest number of rescues were on Mt. San Antonio and at Aztec Falls and
Big Falls. The Gi
*
score at Big Falls is 73.1, the Gi
*
score at Aztec Falls is 58.9, and the Gi
*
score
on Mt. San Antonio is 31.4. Because it encompasses a larger area than the neighborhood
considered in the calculation, the Gi
*
score on Mt. San Antonio is lower than that of Aztec Falls
and Big Falls, although eight rescues were performed there in May, while Aztec Falls had five
and Big Falls had seven.
Figure 15 shows the hot spots and kernel densities for June, July, and August, the busiest
months for hoist rescues. Forty-eight percent of the rescues in the top six areas occurred in these
months. In June the greatest number of rescues were at Mt. San Antonio, Aztec Falls, and Big
Falls. The Gi
*
score for Big Falls is 92.1, the Gi
*
score for Aztec Falls is 88.6, and the Gi
*
score
for Mt. San Antonio is 39.2.
56
Figure 15 Kernel densities and hot spots for June, July, and August.
57
Once again, Mt. San Antonio had more rescues (16) than Aztec Falls (13) and Big Falls (12), but
a lower Gi
*
score.
In July the greatest number of rescues were on Mt. San Antonio and at Bonita Falls,
Aztec Falls, and Big Falls. The Gi
*
score for Aztec Falls is 99.1, the Gi
*
score for Bonita Falls is
76.5, and the Gi
*
score for Big Falls is 64.8. The Gi
*
score for Mt. San Antonio is 30.3, although
12 rescues were recorded there, in comparison to 11 at Big Falls.
In August the highest number of rescues were on Mt. San Antonio and at Bonita Falls,
Deep Creek Hot Springs, and Big Falls. The Gi
*
score for Bonita Falls is 83.5 and the Gi
*
score
for Big Falls is 69.6. The the Gi
*
score for Mt. San Antonio is 59.7, and the Gi
*
score for Deep
Creek Hot Springs is 36.8. Mt. San Antonio had the greatest number of rescues in August with
22, but its Gi
*
score is lower than that of Bonita Falls and Big Falls.
Figure 16 shows the hot spots and kernel densities for September, October, and
November. There were fewer hoist rescues during these months than the summer. In September
the highest Gi
*
scores are in the Devore foothills, which is not one of the top six locations, and
Big Falls. The Gi
*
score for Big Falls for September is 84.8 and the Gi
*
score for the Devore
foothills is 83.2. The high number of rescues in the Devore foothills in September can be traced
to a rescue on September 30, 2010, when eight victims were rescued.
A high kernel density can be seen in October in Siberia Creek, which is not one of the top
six locations. The high number of hoist rescues in October in the Siberia Creek area can be
traced to a rescue of three victims on October 12, 1998 and rescues of three and two victims on
58
Figure 16 Kernel densities and hot spots for September, October, and November.
59
October 16 and 17, 2014. The Gi
*
score for Siberia Creek for October is 57.3. Big Falls and
Bonita Falls both have Gi
*
scores of 66.8, although they had only five rescues each in October.
In November the greatest number of rescues was on Mt. San Antonio. Twenty-one
rescues were recorded, which is its second highest monthly total after August. This number can
be traced to rescues of four victims on November 27, 2005 from Devil’s Backbone, four on
November 27, 2010 from the Sierra Hut, and six on November 17, 2014 from the Sierra Hut. The
Gi
*
score for Mt. San Antonio in November is 105.
Figure 17 shows the hot spots and kernel densities for December, January, and February.
December is a slow month for hoist rescues. The greatest number of rescues were on Mt. San
Antonio and in Icehouse Canyon. Mt. San Antonio had eight rescues and Icehouse Canyon had
6, but they have the same Gi
*
score of 70.6.
January has the third highest number of hoist rescues for a month on Mt. San Antonio
with 20. The Gi
*
score for Mt. San Antonio is 145. Three rescues recorded in January on Mt. San
Antonio had two victims and the rest had one, so the number was not inflated by rescues with
multiple victims and reflects a large number of separate incidents. The large number of rescues
in January on Mt. San Antonio is probably the reason that January has the fourth highest number
of rescues above 6,000 ft., as seen in Figure 10.
The greatest number of rescues in February were on Mt. San Antonio, with eight, and in
Icehouse Canyon, with 14. The Gi
*
score for Mt. San Antonio is 63, and the Gi
*
score for
Icehouse Canyon is 104. The number in Icehouse Canyon can be traced to two rescues on
February 6, 2006 with four and five victims.
60
Figure 17 Kernel densities and hot spots for December, January, and February.
61
The monthly hot spot and density calculations, along with the charts of rescue statistics,
provide a record of the hoist rescue activity at the top six locations throughout the year. The side-
by-side comparison of the hot spot and density calculations with the rescue statistics shows the
importance of comparing locations with similar sizes in a hot spot calculation. Because they
cover more area than the second-order adjacency network used for the calculation, rescues
recorded as happening in the areas designated as Mt. San Antonio or Icehouse Canyon may not
contribute to the Gi
*
score for the same target grid cell. As a result, the Gi
*
scores for Mt. San
Antonio and Icehouse Canyon are lower than the other areas for the same number of rescues.
The size of the cluster could not be increased, however, because a larger cluster would allow
neighboring canyons to contribute to each other and produce imprecise results. Subdividing Mt.
San Antonio and Icehouse Canyon into smaller locations the size of a second-order adjacency
network would give the named locations consistent size, which would give the hot spot
calculations a consistent basis and make the Gi
*
scores more comparable with each other.
Smaller locations, however, would not be consistent with the usage of the names of these areas
within the hiking community or sheriff’s aviation.
As demonstrated in this section, the hoist rescue database allows the incidents behind
unexpected monthly hoist rescue statistics to be investigated. In many cases, a high density in an
area can be traced to a small number of incidents in which multiple victims were rescued.
62
Chapter 5 Conclusion
The main conclusion that can be drawn from this project is that the aviation unit’s records,
although their original purpose was equipment maintenance and accounting, contain valuable
geographic information about its operations. The combination of hoist log books, blue cards, and
personal records of unit members allowed the locations of rescues to be determined precisely
enough to analyze trends in the elevation of hoist rescues and identify the areas of highest
concentration of rescues and seasonal trends in those areas. The process of georeferencing and
digitizing the records was labor intensive, but the results can contribute to discussions about the
history, current status, and future of the hoist rescue program in the sheriff’s department. Section
5.1 discusses the trends in the elevations and locations of hoist rescues. Section 5.2 discusses
potential applications and analysis possibilities for the database, and section 5.3 discusses other
ways in which GIS can be applied to the operations of the aviation unit. Finally, section 5.4
discusses the potential for search and rescue agencies to use GIS to learn from their historical
records and improve their operations.
5.1. Trends in Hoist Rescues
Mt. San Antonio, Icehouse Canyon, Bonita Falls, Deep Creek Hot Springs, Aztec Falls, and
Big Falls share characteristics that result in frequent hoist rescues. They are exciting locations
that attract large numbers of people, they have hazardous terrain that can result in injuries, they
pose challenges for ground teams to move injured victims, and there are few flat, clear areas
large enough to land a helicopter close to where injuries occur. The identification of these
locations and the seasonal trends in rescue activity will probably not come as a surprise to
experienced members of the aviation unit. The results will, however, provide objective evidence
for the anecdotal knowledge of unit members and a basis for discussions with people interested
63
in but unfamiliar with the team’s operations, such as the leadership of the sheriff’s department,
officials in the San Bernardino County government, and new members of the team. The
personnel in the aviation unit is constantly changing due to retirements, promotions, and changes
in assignment. A record of hoist records can make it easy to bring new members up to speed with
this aspect of the unit’s activities and understand how the equipment is used. The elevation
trends in figure 9, for example, can be used to build a case for the necessary capabilities of a new
rescue helicopter for work in the county.
The seasonal trends in hoist rescues generally show that Mt. San Antonio experiences a
high number of rescues in the summer and winter, Icehouse Canyon is busy in the late winter,
and the three waterfalls see their greatest activity in the summer. The results also show, however,
that hoist rescues can occur in these areas at unexpected times. In addition, the top six areas
account for almost 40% of rescues, so the remaining 60% of rescues occur in other parts of the
county and surrounding areas. The analysis of the rescue records can help anticipate rescue
activity and focus readiness and prevention efforts in certain areas, but the aviation unit must
maintain the ability to respond to a rescue anywhere in the county at any time.
The results of this project can be used to direct efforts to prevent injuries in the areas
where the most hoist rescues happen. The ground search and rescue team whose area includes
Mt. San Antonio, Icehouse Canyon, and Bonita Falls, for example, could use this information to
anticipate rescues or plan patrols on Mt. San Antonio in January and Icehouse Canyon in
February. The team could also plan patrols on Mt. San Antonio and Bonita Falls in the summer.
The Forest Service has closed Aztec Falls and the area above the first waterfall at Big
Falls to visitors. Based on the seasonal trends in hoist rescues in those locations before the
closures, it can be recommended that the Forest Service should focus its efforts in maintaining
64
barriers and signs and patrolling the trailheads to keep people out of the closed areas in June,
July, and August.
5.2. Hoist Rescue Database Analysis Possibilities
There are many possibilities for analysis using the hoist rescue database beyond what was
carried out in this project, and possibilities to improve the documentation so that in the future, no
records have to be discarded, there is no uncertainty as to the locations of rescues, and more
pertinent information is available for analysis. Sections 5.2.1 and 5.2.2 describe additional
applications of the current database and section 5.2.3 describes ways to augment the rescue
documentation to collect more useful information.
5.2.1. Rescue Database Analysis Possibilities
The hot spot and kernel density calculations described in Chapters 3 and 4 used the
number of victims as the weighting attribute. In several of the monthly calculations, a high Gi
*
score and density could be traced to a small number of incidents in which multiple victims were
rescued. These results reflect the number of people that were rescued, but if the leadership of the
unit preferred to use the number of incidents as a measure of rescue activity, it would be a simple
task to recalculate the hot spots and kernel densities with the number of incidents and no
weighting attribute.
The rescue database records the aircraft that was used in each hoist rescue. This attribute
was used to compare the elevations of rescues performed by the patrol and medium helicopters,
as shown in figure 10, but was not used in the hot spot or kernel density calculations. Air Rescue
306 and 307 have approximately the same power, but 307 weighs almost 1000 lbs. more than
306. The documentation for this attribute is not perfect, as 235, or 31%, of the 761 hoist rescues
performed by the medium helicopters do not specify the aircraft. However, it may be interesting
65
to compare the elevations of the rescues performed by 306 and 307 to see if the additional weight
has made a difference in the rescues the helicopters have performed. Air Rescue 308 has only
performed nine hoist rescues since it was equipped with a hoist, but as it continues to be used as
a hoist ship, the locations, elevations, and temperatures of its rescues can be used to refine the
assessment of its capabilities.
Another attribute of the rescue database that was found in the narrative section of some
hoist records and blue cards is whether the rescue was assisted by a helitac-trained ground team.
In these rescues, the aircrew relied on personnel from a fire or police department, the California
Highway Patrol, or a volunteer search and rescue team to perform an aspect of the hoist rescue,
usually packaging the victim in a rescue device and connecting the hoist hook. The locations
where these rescues have taken place and agencies involved in assisting the rescues can help the
aviation unit determine where and with whom to conduct joint trainings. Training with other
agencies allows their personnel to become familiar with the rescue procedures, hand and arm
signals, equipment, and radio frequencies used by the aviation unit, and will improve the
coordination and safety of the rescues they perform together.
The aviation unit has used night vision goggles (NVGs) for years for patrol purposes, and
has recently begun performing hoist rescues at night with them. Establishing and maintaining the
capability of pilots and crew chiefs to perform hoist rescues with NVGs requires a significant
amount of training. Tracking and analyzing the trends in night rescues can help demonstrate the
need for this capability and justify the investment of time and money to maintain it.
5.2.2. Application Dashboard
Another possibility for the rescue database would be to create an application dashboard to
make the information available for analysis via the internet. A web GIS could generate maps and
66
figures based on the results of searches by date, location, elevation, person, or any other attribute
of the database. The dashboard would be useful for the leadership of the aviation unit, who could
use it to generate maps and charts similar to the ones in Figures 11 and 12 for use in their
planning and discussions with sheriff’s department and San Bernardino County officials. The
dashboard would also be an excellent way to inform the public about the rescues the aviation unit
performs and allow users to generate maps based on their own interests. The application
dashboard would also be a good way to drive recruitment and appeal for donations.
Current and former aviation unit members would probably enjoy using the application
dashboard to generate maps and statistics of the rescues they participated in and compare them to
others. Although it would not analyze or apply directly to rescue operations, the ability to see
maps and charts documenting their rescues would be a morale boost. The most important
component of the aviation unit is the people, and the members of the air rescue team are
volunteers. The application dashboard would be a way to recognize their contributions and honor
their legacy by making the most of the information gathered about their rescues.
5.2.3. Information Gathering
The process of georeferencing the hoist rescues and the analysis that was carried out
using the database can help determine what information should be collected about future rescue
operations and how the documentation forms and processes should be changed. Recording GPS
coordinates for each rescue would eliminate the georeferencing process and the need to discard
records when their locations cannot be precisely determined from text descriptions. Obtaining
GPS coordinates would be a trivial matter for the patrol helicopters, because they have an
integrated navigation system and a large map display in their dashboards. The medium
helicopters do not have integrated navigation systems, and the pilot and crew chief may be too
67
busy to record a location using a handheld GPS device during a hoist rescue, when they have
other procedures to follow. If they were collected, though, GPS coordinates would precisely
determine the location and elevation of each rescue, which would improve the hot spot and
density calculations of sections 4.2 and 4.3 and allow other types of analysis to be carried out.
Recording GPS coordinates would eliminate the uncertainty in the locations of the hoist
rescues and allow them to be recorded as points rather than as polygons. To conduct spatial
autocorrelation and hot spot calculations, however, the rescues would still have to be aggregated
over a grid. In this thesis project, the size of the grid cells was dictated by the area of the rescue
location polygons, and was chosen to reflect the uncertainty of the rescue locations while
preserving as much spatial detail as possible. If the rescue locations were recorded as exact
points, grid cells of any size could be used. It would be interesting to explore the MAUP by
comparing the memory, computer time, and results of statistical calculations for different
aggregation schemes in a fashion similar to the references in sections 2.2 and 2.3.
Another useful piece of information that is currently not regularly being collected is the
outside air temperature. As was discussed in section 1.5, the temperature has a large effect on
how helicopters function. The outside air temperature, along with the elevation, would allow the
density altitude to be calculated for each rescue, which would help build understanding of the
capabilities of the helicopters and inform pilots and crew chiefs when a rescue is approaching
their limits and measures to reduce weight should be taken.
The database developed in this project records only the rescues the aviation unit has
successfully handled, and thus reflects a “survivor bias.” To understand the limits of the
helicopters, it would be useful to have records of rescues they were unable to perform with an
explanation of whether the rescue was prevented by visibility, elevation, weather conditions, or
68
other factors. The dispatch system may be a source for information about instances when the
helicopter was requested but was unable to complete the rescue. Analyzing how many rescues
the aviation unit was unable to complete and the reasons they were unable to complete them
could define its limiting factors and build a case for how many more rescues would likely be
performed with more capable helicopters or different training requirements for crew members.
To streamline the process of collecting data about future hoist missions, a database could
be established using a program such as Microsoft Access (Microsoft 2020) with an interface for
users to create and add records. If updating this database became a standard procedure for
documenting rescues, the records could be used to generate blue cards and eliminate the need for
the hoist log books. The geographic coordinates and other attributes of the records could be
exported to the web GIS discussed in section 5.2.2 to keep it up to date.
5.3. Other Applications of GIS for Air Rescue
There are many other ways in which GIS can be used to enhance the operations of the
aviation unit beyond recording and analyzing hoist rescues. One mission the rescue helicopters
are regularly tasked with is transporting ground search and rescue team members to areas they
have been assigned to search. Inserting teams by helicopter increases the amount of time they
have to search their assigned areas and reduces their exposure to hazardous terrain.
The sheriff’s Incident Management Team (IMT) is a volunteer team that runs the
command post during large scale search operations (Nicolet 2011). The members use GIS to
record victim information, such as their planned route and place last seen; assign search areas
based on the probable behavior of the victim; and document areas searched and clues found. The
use of GIS in search and rescue has been an active area of research in recent years (Pfau 2018).
Tools developed to predict the behavior of lost hikers have been used in USC master’s thesis
69
projects that analyzed the records of search operations in Oregon (Johnson 2106) and identified
places where people lost their way in Yosemite National Park (Danser 2018). The IMT’s records
of recent large scale searches could be used to identify areas that have been assigned to search
teams in more than one operation. Hikers go missing several times a year on Mt. San Antonio,
for example, and there are probably areas that have been searched as part of more than one
operation. The records could be used to find the search areas that would benefit the most from
having teams inserted by helicopter because they are difficult to access from trailheads.
Within these search areas, GIS could be used to identify large, clear, flat areas as
candidate helicopter landing zones (HLZs). A recent USC master’s thesis developed a method to
use GIS to identify candidate HLZs that volunteer teams could survey in person for use in
disaster relief operations (Darby 2017). A one-meter digital elevation model (DEM) is available
for most of the San Bernardino Mountains from the USGS National Map website (USGS 2020),
and could be used to calculate slope. Land cover data is available from the National Land Cover
Database (MRLC 2020), electrical transmission line data is available from the California Energy
Commission (California Energy Commission 2020), and electrical distribution line data is
available from Southern California Edison (SCE 2020). With these inputs, it would be possible
to perform a site suitability analysis and identify candidate landing sites that would be useful for
search operations.
The candidate sites would have to be surveyed in person to confirm their suitability, then
a list with the names, coordinates, elevations, and descriptions of the sites could be compiled.
The list would be shared with the IMT and aviation unit and would allow the IMT to plan the
insertion of search teams by helicopter.
70
The sites would not be maintained, and using them would be at the discretion of the pilot,
but the list of sites would be an important resource for the IMT in planning search operations.
5.4. GIS Analysis of Rescue Operations
Historical search and rescue records represent an opportunity for agencies to learn from
the operations they have carried out. Retrospective studies of these records can identify trends in
where and when missions happen and help teams determine the best ways to prepare for and
handle them. Even if the records do not include coordinates or were not collected with the intent
of supporting geographic analysis, it is possible, by combining different sources of information,
to georeference the locations of incidents with enough precision to identify spatial and temporal
patterns.
Search and rescue operations often happen at inconvenient times and in uncomfortable
and dangerous conditions. Many of the people who carry them out are volunteers. A very good
way to honor their efforts is to learn as much as possible from these operations to anticipate
future incidents, guide prevention efforts, and improve their teams’ training and readiness.
The precision of the information recorded in search and rescue documentation limits the
analysis that it can support. Improving the data collection process, for example by including
geographic coordinates and automatically updating a database, can ensure that the greatest
possible benefit is realized from operations in the future, so that search and rescue incidents can
be prevented, and teams can be as prepared as possible to handle them when they do occur.
71
References
Airbus. 2020. “H125.” Airbus. https://www.airbus.com/helicopters/civil-
helicopters/intermediate-single/h125.html. Accessed August 6, 2020.
Aircraft Owners and Pilots Association (APOA). 2020. “Density Altitude.” Aircraft Owners and
Pilots Association. https://www.aopa.org/training-and-safety/active-pilots/safety-and-
technique/weather/density-altitude. Accessed October 29, 2020.
Amberg, Brian. 2021. “The quartic kernel.” Wikipedia.
https://en.wikipedia.org/wiki/Kernel_(statistics)#/media/File:Kernel_quartic.svg.
Accessed May 1, 2021.
Anselin, Luc. 1995. “Local Indicators of Spatial Autocorrelation – LISA.” Geographical
Analysis 27, no. 2: 93-115.
Araza, Adam. 2019. “California Ballot Results Viewer: 2008-2018 A Web GIS Application for
Viewing Ballot Proposition Results in California.” Master’s Thesis, University of
Southern California.
California Energy Commission. 2020. “California Electric Transmission Line.” California
Energy Commission. https://cecgis-caenergy.opendata.arcgis.com/datasets/california-
electric-transmission-line. Feb 7, 2020.
California Legislative Information. 2020. “ARTICLE 1. Duties [26600 - 26616].” California
Legislative Information.
http://leginfo.legislature.ca.gov/faces/codes_displayText.xhtml?lawCode=GOV&division
=2.&title=3.&part=3.&chapter=2.&article=1. Accessed November 16, 2020.
CMC. 2020. “Helitack Hotseat.” CMC Rescue, Inc.
https://www.cmcpro.com/equipment/helitack-hotseat/. Accessed August 6, 2020.
CMC. 2020. “Lifesaver Victim Harness.” CMC Rescue, Inc.
https://www.cmcpro.com/equipment/lifesaver-victim-harness/. Accessed August 6, 2020.
CMC. 2020. “Titanium Split-Apart Rescue Litter.” CMC Rescue, Inc.
https://www.cmcpro.com/equipment/titanium-split-apart-rescue-litter/. Accessed August
6, 2020.
Danser, Raymond A. 2018. “Applying Least Cost Path Analysis to Search and Rescue Data: A
Case Study in Yosemite National Park.” Master’s Thesis, University of Southern
California.
Darby II, Keith Caldwell. 2017. “Developing, Maintaining, and Employing Crowd-sourced
Geospatial Data in Support of Helicopter Landing Zone Surveys for Disaster Response
Operations.” Master’s Thesis, University of Southern California.
72
Doherty, Paul J., Yu Liu, Jared Doke, Qinghua Guo, and John 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.
Doherty, Paul J., Qinghua Guo, and Otto Alvarez. 2013. "Expert Versus Machine: A Comparison
Of Two Suitability Models For Emergency Helicopter Landing Areas In Yosemite
National Park.” The Professional Geographer 65, no. 3: 466-481.
Doherty, Paul J., Qinghua Guo, Wenkai Li, and Jared Doke. 2014. “Space-Time Analyses for
Forecasting and Understanding Future Incident Occurrence: A Case-Study from
Yosemite National Park Using the Presence and Background Learning Algorithm.”
International Journal of Geographical Information Science 28, no. 5: 910-927.
Doherty, Paul J., Quinghua Guo, Jared Doke, and Don Ferguson. 2014. “An Analysis of
Probability of Area Techniques for Missing Persons in Yosemite National Park.” Applied
Geography 47: 99–110.
Environmental Systems Research Institute (Esri). 2020. “What is GIS?” Esri.
https://www.esri.com/en-us/what-is-gis/overview. Accessed July 17, 2020.
Foo, Cheryl P. Z., Mahvareh Ahghari, and Russell D. MacDonald. 2010. "Use Of Geographic
Information Systems To Determine New Helipad Locations And Improve Timely
Response While Mitigating Risk Of Helicopter Emergency Medical Services
Operations.” Prehospital Emergency Care 14, no. 4: 461-468.
Getis, Arthur and J. K. Ord. 1992. “The Analysis of Spatial Association by Use of Distance
Statistics.” Geographical Analysis 24, no. 3: 189-206.
Goodrich Hoist and Winch. 2020. “About Us.” Collins Aerospace.
https://www.goodrichhoistandwinch.com/company/about/. Accessed on October 14,
2020.
Goodrich Hoist and Winch. 2020. “RESCUE HOIST 42305.” Collins Aerospace.
https://www.goodrichhoistandwinch.com/products/detail/rescue-hoist-42305/. Accessed
on August 6, 2020.
Goodrich Hoist and Winch. 2020. “RESCUE HOIST 44301.” Collins Aerospace.
https://www.goodrichhoistandwinch.com/products/detail/rescue-hoist-44301/. Accessed
on August 6, 2020.
Gundran, Robert. 2019. “Off-trail area of Big Falls Trail in Forest Falls to be closed to public
until May 2020.” The Sun. https://www.sbsun.com/2019/07/03/off-trail-area-of-big-falls-
trail-in-forest-falls-to-be-closed-to-public-until-may-2020/. Accessed November 5, 2020.
Hornberger, Zachary T., Bruce A. Cox, and Raymond R. Hill. 2019. “Effects of Aggregation
Methodology on Uncertain Spatiotemporal Data.” Preprint. Geographical Analysis.
https://arxiv.org/abs/1910.05125. Accessed June 15, 2021.
73
Johnson, Mark Powell. 2016. “Evaluating the Utility of a Geographic Information Systems-based
Mobility Model in Search and Rescue Operations.” Master’s Thesis, University of
Southern California.
Kalinic, Maja and Jukka M. Krisp. 2018. “Kernel Density Estimation (KDE) vs. Hot-Spot
Analysis - Detecting Criminal Hot Spots in the City of San Francisco.” In Geospatial
Technologies for All: short papers, posters and poster abstracts of the 21th AGILE
Conference on Geographic Information Science, edited by Mansourian, A., P. Pilesjö, L.
Harrie, and R. von Lammeren. Lund, Sweden: AGILE. 12-15 June 2018.
Koester, Robert J. 2008. Lost Person Behavior. Charlottesville: dbS Productions LLC.
Kroh, P. 2020. “Identification of Landing Sites for Rescue Helicopters in Mountains with use of
Geographic Information Systems.” Journal of Mountain Science 17, no. 2: 261-270.
Manwaring, Jan C., George A. Conway, and Larry C. Garrett. 1998. “Epidemiology and
Prevention of Helicopter External Load Accidents.” Journal of Safety Research 29, no. 2:
107–121.
Mathews, A. J., and E. A. Ellis. 2016. “An evaluation of tornado siren coverage in Stillwater,
Oklahoma: Optimal GIS methods for a spatially explicit interpretation.” Applied
Geography 68: 28–36.
Microsoft. 2020. “Elevate Data.” Microsoft. https://www.microsoft.com/en-us/microsoft-
365/access. Accessed December 7, 2020.
Multi-Resolution Land Characteristics Consortium (MRLC). 2020. “NLCD 2016 Land Cover
(CONUS).” MRLC.
https://www.mrlc.gov/data?f%5B0%5D=category%3ALand%20Cover. Accessed July
12, 2020.
Nelson, J. K., and C. A. Brewer. 2017. “Evaluating data stability in aggregation structures across
spatial scales: revisiting the modifiable areal unit problem.” Cartography and
Geographic Information Science 44 no. 1: 35–50.
Nicolet, Dave. 2011. “Incident Management Team.” San Bernardino Sheriff’s SAR.
http://sbsar.org/news-information/2011/4/13/incident-management-team.html. Accessed
November 17, 2020.
Openshaw, Stan. 1983. The modifiable areal unit problem. Norwick: Geo Books.
Pfau, Loren and Justine I. Blanford. 2018. “Use of Geospatial Data and Technology for
Wilderness Search and Rescue by Nonprofit Organizations.” The Professional
Geographer 70, no. 3: 434-442.
San Bernardino County. 2015. “Community Indicators Report 2015.” San Bernardino County.
http://cms.sbcounty.gov/Portals/21/Resources%20Documents/CIR_2015_Report.pdf.
Accessed November 23, 2020.
74
San Bernardino County, CA Code of Ordinances. 2019. “CHAPTER 5: SHERIFF
DEPARTMENT.” American Legal Publishing.
http://library.amlegal.com/nxt/gateway.dll/California/sanbernardinocounty_ca/sanbernard
inocountycaliforniacodeofordin?f=templates$fn=default.htm$3.0$vid=amlegal:sanbernar
dinocounty_ca. Accessed November 16, 2020.
San Bernardino County Fire. 2020. “CONFIRE Dispatch.” County of San Bernardino.
https://www.sbcfire.org/Programs/CONFIREDispatch.aspx. Accessed October 23, 2020.
San Bernardino County Sheriff’s SAR. 2020. “sbsar.org.” Inland Empire Search and Rescue
Council. http://sbsar.org/. Accessed August 6, 2020.
San Bernardino County Sheriff’s Department. 2020. “Air Medics.” San Bernardino County
Sheriff’s Department. https://wp.sbcounty.gov/sheriff/air-medics/. Accessed August 6,
2020.
Schappert, Jason. 2020. “Calculating Density Altitude.” MzeroA.com.
https://www.m0a.com/wordpress/wp-content/uploads/2010/11/Density-Altitude-
Chart.jpg. Accessed October 29, 2020.
Schnebele, E. and G. Cervone. 2013. “Improving Remote Sensing Flood Assessment Using
Volunteered Geographical Data.” Natural Hazards and Earth Systems Science 13: 669–
677.
Southern California Edison (SCE) – Document Library. 2020. “SCE High Fire Area Emergency
Documents – San Bernardino County.” Southern California Edison.
https://library.sce.com/?10000_group.propertyvalues.property=jcr%3Acontent%2Fmetad
ata%2Fcq%3Atags&10000_group.propertyvalues.operation=equals&10000_group.prope
rtyvalues.0_values=sce-document-library%3Asce-high-fire-area-emergency-
documents%2Fsan-bernardino-county%2Fgis. Accessed June 1, 2020.
United States Census Bureau. 2019. “2019 TIGER/Line® Shapefiles: Counties (and
equivalent).” US Department of Commerce. https://www.census.gov/cgi-
bin/geo/shapefiles/index.php?year=2019&layergroup=Counties+%28and+equivalent%29
. Accessed July 24, 2020.
United States Census Bureau. 2019. “2019 MAF/TIGER Feature Class Codes.” US Department
of Commerce. https://www2.census.gov/geo/pdfs/reference/mtfccs2019.pdf. Accessed
October 22, 2020.
United States Census Bureau. 2019. “2019 TIGER/Line® Shapefiles: Roads.” US Department of
Commerce. https://www.census.gov/cgi-
bin/geo/shapefiles/index.php?year=2019&layergroup=Roads. Accessed October 22,
2020.
United States Forest Service (USFS). 2020. “Aztec Falls and surrounding areas of Deep Creek to
temporarily close.” United States Department of Agriculture.
75
https://www.fs.usda.gov/detail/sbnf/news-events/?cid=FSEPRD743425. Accessed
November 5, 2020.
United States Forest Service (USFS). 2020. “Big Falls Trail 1E13.” United States Department of
Agriculture.
https://www.fs.usda.gov/recarea/sbnf/recarea/?recid=26557#:~:text=Upper%20Big%20F
alls%20is%20currently,up%20to%20the%20waterfalls%20itself. Accessed November
17, 2020.
United States Geological Survey (USGS). 2020. “The National Map.” United States Department
of the Interior. https://viewer.nationalmap.gov/basic/. Accessed July 12, 2020.
van Wijk, Jarke J. 2005. “The Value of Visualization.” IEEE Visualization 2005.
Wieczorek, John, Qinghua Guo, and Robert Hijmans. 2004. “The Point-Radius Method for
Georeferencing Locality Descriptions and Calculating Associated Uncertainty.”
International Journal of Geographical Information Science 18, no. 8: 745-767.
Xu, P., H. Huang, N. Dong, and M. Abdel-Aty. 2014. “Sensitivity analysis in the context of
regional safety modeling: Identifying and assessing the modifiable areal unit problem.”
Accident Analysis & Prevention 70: 110–120.
Abstract (if available)
Abstract
People become lost and injured in remote areas on a daily basis. Search and rescue personnel, including members of the military, law enforcement agencies, fire departments, emergency medical services, and volunteer teams stand by for the opportunity to rescue someone. This requires the development of teamwork and skills and the acquisition and maintenance of equipment. Rapid and accurate analysis of geographic information plays a critical role in ensuring resources are used as safely and effectively as possible in search and rescue operations. ❧ Historical search and rescue records contain a wealth of information that can guide training, drive recruitment of new members, and define the necessary capabilities of equipment and personnel. Although they may not have been collected with the intent of providing data for spatial analysis, the geographic information contained in rescue documentation can provide information as to the trends in where rescues are occurring, which can help anticipate future incidents and improve readiness. ❧ This thesis is a retrospective application of geographic information science to the records of the San Bernardino County sheriff’s air rescue program. The hoist rescues performed since the program’s inception in 1993 were georeferenced, digitized, and used as the basis for a geodatabase. The spatial pattern of air rescues in the county was analyzed to identify the areas of greatest rescue activity and seasonal trends in locations and elevations. This information will help leaders on the team guide training and preparation and help define the necessary capabilities of rescue helicopters for work in the county. In addition, by showing trends in rescue activity in certain areas and times of the year, recommendations are made for preventative measures to keep hikers and climbers safe. The analysis methods here can be applied to similar projects throughout the United States and beyond.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
A spatial and temporal exploration of how satellite communication devices impact mountain search and rescue missions in California’s Sierra Nevada mountain range
PDF
A spatiotemporal analysis of environmental risk factors of Lyme disease in the Northeastern United States
PDF
Enriching the Demographic Survey sampling for the Los Angeles County Annual Homeless Count with spatial statistics
PDF
Happy Traveler: discovering joy on university campuses and beyond through a web-based GIS application
PDF
A geospatial analysis of income level, food deserts and urban agriculture hot spots
PDF
Tracking trends in earthquakes and tropical storms: a web GIS application
PDF
Site suitability analysis for implementing tidal energy technology in southern California
PDF
Silicon Valley construction project web mapping application
PDF
Providing a new low-cost primary care facility for under-served communities: a site suitability analysis for Service Planning Area 6 in Los Angeles County, California
PDF
An exploration of the spatiotemporal distribution of snow crab (Chionoecetes opilio) in the eastern Bering Sea: 1982 – 2018
PDF
Development of a Web GIS application to aid marathon runners in the race selection and planning process
PDF
Using pedestrian accessibility indicators to locate schools: a site suitability analysis in Greenville County, South Carolina
PDF
Collecting and managing VGI infrastructure assessments in support of stability operations
PDF
The role of precision in spatial narratives: using a modified discourse quality index to measure the quality of deliberative spatial data
PDF
Using the Digital Shoreline Analysis System (DSAS) to analyze changes in shoreline position caused by seawalls along a section of Oregon's coast
PDF
A spatial analysis of veteran healthcare accessibility
PDF
Rails to trails web mapping application for the great Redwood trails: mapping northern California’s repurposed trails
PDF
Using GIS to predict human movement patterns in complex humanitarian emergencies: a test case of the Syrian Conflict
PDF
An exploratory spatial analysis of fire service and EMS accessibility in northeastern Illinois communities
PDF
Analysis of park accessibility in Redan, Georgia Web GIS application
Asset Metadata
Creator
Fisher, Daniel Ross
(author)
Core Title
GIS analysis of helicopter rescue in San Bernardino County, California
School
College of Letters, Arts and Sciences
Degree
Master of Science
Degree Program
Geographic Information Science and Technology
Degree Conferral Date
2021-12
Publication Date
09/30/2021
Defense Date
06/23/2021
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
georeferencing,helicopter rescue,hot spot analysis,kernel density,OAI-PMH Harvest,rotary wing rescue,San Bernardino County
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Bernstein, Jennifer (
committee chair
), Fleming, Steven (
committee member
), Sedano, Elisabeth Jane (
committee member
)
Creator Email
danielrossfisher@gmail.com,fisherda@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-oUC16011560
Unique identifier
UC16011560
Legacy Identifier
etd-FisherDani-10116
Document Type
Thesis
Format
application/pdf (imt)
Rights
Fisher, Daniel Ross
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the 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. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.
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
Tags
georeferencing
helicopter rescue
hot spot analysis
kernel density
rotary wing rescue