University of Southern California Dissertations and Theses

Spatial analysis of human activities and wildfires in the Willamette National Forest

(USC Thesis Other)

Spatial analysis of human activities and wildfires in the Willamette National Forest

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Spatial Analysis of Human Activities and Wildfires in the
Willamette National Forest

by

William Dickey

A Thesis Presented to the
Faculty of the USC Graduate School
University of Southern California
In Partial Fulfillment of the
Requirements for the Degree
Master of Science
(Geographic Information Science and Technology)

December 2018

ii

Copyright © 2018 by William Dickey

iii

To my wife Sarah, and the Twigglers; Zion and Asher

iv
Table of Contents
List of Figures ............................................................................................................................... vii
List of Tables ................................................................................................................................. ix
List of Abbreviations ..................................................................................................................... xi
Abstract ......................................................................................................................................... xii
Chapter 1 Introduction .................................................................................................................... 1
1.1. Wildfires and the Willamette National Forest ....................................................................3
1.1.1. History of Wildfires in the Willamette National Forest.............................................3
1.2. Common Activities Found in the Willamette National Forest ...........................................4
1.2.1. Recreational Activities in the Willamette National Forest ........................................5
1.2.2. Industrial Activities in the Willamette National Forest .............................................5
1.3. Wildfire Impacts on Economic Factors ..............................................................................6
1.4. Project Objectives ...............................................................................................................7
1.5. Organization of this Document ...........................................................................................8
Chapter 2 Related Research ............................................................................................................ 9
2.1. Human Activities Related to Wildfire Occurrences ...........................................................9
2.1.1. Relationship of Recreational Activities to Wildfire Occurrences ............................10
2.1.2. Relationship of Human Infrastructure to Wildfire Ignitions....................................11
2.1.3. Relationship of Industrial Activities to Wildfire Occurrences ................................12
2.2. Wildfire Prevention Through Forest Management ...........................................................13
2.3. The Role of Weather in Wildfire Occurrence ...................................................................14
2.3.1. Fair Weather Activities and Impact on Fire Frequency ...........................................15
2.4. Modeling Wildfires using Maximum Entropy Techniques ..............................................16
2.5. Summary ...........................................................................................................................17
v
Chapter 3 Data and Preliminary Data Exploration ....................................................................... 18
3.1. Study Area: Willamette National Forest ...........................................................................18
3.1.1. Geographical Features of The Willamette National Forest .....................................18
3.1.2. Recreation in the Willamette National Forest ..........................................................21
3.1.3. Vegetation ................................................................................................................22
3.1.4. Climate .....................................................................................................................22
3.1.5. The Role of the Forest Service and Other Agencies ................................................23
3.2. Data Overview ..................................................................................................................24
3.2.1. Forest Boundary Data ..............................................................................................25
3.2.2. Wildfire Data ...........................................................................................................25
3.2.3. Recreational Site and Human Infrastructures Data ..................................................29
3.2.4. Road Data.................................................................................................................34
3.2.5. Trail Data .................................................................................................................36
3.2.6. Private, Residential, and Lodging Areas..................................................................36
3.2.7. Administrative Areas ...............................................................................................37
3.2.8. Town Sites ...............................................................................................................38
3.2.9. Summary of Data Layers Used in the Analyses ......................................................39
3.3. Preliminary Data Exploration ...........................................................................................41
3.3.1. Kernel Density .........................................................................................................41
3.3.2. Results of Kernel Density Preliminary Data Exploration ........................................42
3.3.3. Overall Observations from Density Analysis ..........................................................46
Chapter 4 Data Mining with Maxent ............................................................................................ 47
4.1. Using Maxent to Understand Relationships Between Fires and Sites ..............................47
4.2. Categorical Data Preparation ............................................................................................48
4.2.1. Wildfire Data ...........................................................................................................48
vi
4.2.2. Forest Boundary Data ..............................................................................................49
4.2.3. Recreational Site Data..............................................................................................49
4.2.4. Polyline Data: Road and Trail..................................................................................49
4.2.5. Polygon Data: Private/residential, Town, and Administrative Areas ......................50
4.3. Continuous Data Preparation ............................................................................................50
4.4. Maxent Data Methods and Procedures .............................................................................51
4.4.1. Presence and Environmental Variables Input Data .................................................52
Chapter 5 Maxent Results ............................................................................................................. 53
5.1. Categorical Results ...........................................................................................................53
5.2. Continuous Data Results ...................................................................................................53
5.2.1. Campfire Caused Wildfire Model ............................................................................55
5.2.2. Smoking Caused Wildfires ......................................................................................59
5.2.3. Total Wildfires for the Willamette National Forest .................................................63
Chapter 6 Discussion and Conclusion .......................................................................................... 67
6.1. Future Work ......................................................................................................................68
6.2. Fires and Park Users .........................................................................................................68
References ..................................................................................................................................... 70
Appendix A. Locations of Individual Wildfire Causes ................................................................ 75
Appendix B: Recreational Sites and Human Infrastructures ........................................................ 77

vii
List of Figures
Figure 1. The Study Area: The Willamette National Forest, major roads, towns and
communities, and nearby major cities. .................................................................................... 2
Figure 2. Locations of human caused wildfire locations, locations of lightning caused wildfire
occurrences, and combined human and lightning caused wildfire occurrences. ..................... 3
Figure 3. The Willamette National Forest. ................................................................................... 19
Figure 4. Location of the 493 human caused wildfires in the WNF between 1995 and 2008. ..... 26
Figure 5. Flowchart showing the Recsite and human infrastructures data preparation ................ 31
Figure 6. Road data editing process .............................................................................................. 35
Figure 7. Locations of all recreational sites and human infrastructures included in this study.. .. 40
Figure 8. Density of campfire caused wildfires ............................................................................ 43
Figure 9. Density of smoking caused wildfires ............................................................................ 44
Figure 10. Density of human caused wildfires ............................................................................. 45
Figure 11. Maxent model response curves for campfires caused wildfires. ................................. 57
Figure 12. Predicted campfire ignitions ........................................................................................ 58
Figure 13. Maxent model response curves for smoking caused wildfires. ................................... 61
Figure 14. Predicted smoking ignitions. ....................................................................................... 62
Figure 15. Maxent model response curves for smoking caused wildfires .................................... 64
Figure 16. Predicted wildfire ignitions ......................................................................................... 66
Figure 17. Arson related wildfires ................................................................................................ 75
Figure 18. Children related wildfires ............................................................................................ 75
Figure 19. Campfire related wildfires ........................................................................................... 75
Figure 20. Debris Burning related wildfires ................................................................................. 75
Figure 21. Railroad related wildfires ............................................................................................ 76
Figure 22. Equipment use relate wildfires .................................................................................... 76
viii
Figure 23. Miscellaneous related wildfires ................................................................................... 76
Figure 24. Smoking related wildfires............................................................................................ 76
Figure 25. Campground ................................................................................................................ 77
Figure 26. Day Use Area .............................................................................................................. 77
Figure 27. Trail shelter .................................................................................................................. 77
Figure 28. Snowpark ..................................................................................................................... 77
Figure 29. Boating areas ............................................................................................................... 78
Figure 30. Towns and communities .............................................................................................. 78
Figure 31. Adminstrative areas ..................................................................................................... 78
Figure 32. Roads ........................................................................................................................... 78
Figure 33. Residential/private ....................................................................................................... 79
Figure 34. Trails ............................................................................................................................ 79

ix
List of Tables
Table 1. National Interagency Fire Center costs for the years 1995 and 2008 for wildfire
suppression. Source: National Interagency Fire Center 2016. ................................................ 7
Table 2. Number of human caused wildfire occurrences by category, 1995-2008 ...................... 27
Table 3. Sources used to create the recreational site and human infrastructures data layers ....... 29
Table 4. Consolidation of Categories and Naming of Data Layers for Recreational Sites and
Human Infrastructures ........................................................................................................... 33
Table 5. Type Buffer Size ............................................................................................................. 34
Table 6. Kernel Density Input Parameters .................................................................................... 42
Table 7. Parameter settings used for Maxent model runs ............................................................. 51
Table 8. Variable contributions for campfire caused wildfires ..................................................... 55
Table 9. Variable contributions for smoking caused wildfires ..................................................... 59
Table 10. Variable contributions for total wildfire in the Willamette National Forest ................. 63

x
Acknowledgements
I am thankful for my wife Sarah, my mother-in-law Vicky, and my entire family for their
patience and understanding while I wrote this thesis. I would like to say thank you to my advisor
Professor Kemp for her patience and assistance. A large thank you to the U.S. Forest Service
employees Matthew Wilson, Glynis Bauer, Matt Peterson, and Don English for their help and
data they provided for this thesis. Last, but not least, a thank you to all my fellow students who I
had the honor of meeting and exploring Catalina Island with.

xi
List of Abbreviations
ASCII American Standard Code for Information Interchange
ATV All-Terrain Vehicle
AUC Area Under Curve
BRT Boosted Regression Tree
CSV Comma-Separated Value
GPS Global Positioning System
INFRA Infrastructure
NAD North American Datum
NW North West
ResPVT Residential/private
ROC Receiver Operating Characteristics
SOPA Schedule Of Proposed Actions
USDA United States Department of Agriculture
USFS United States Forest Service
WNF Willamette National Forest
WUI Wildland Urban Interface

xii
Abstract
Across the nation wildfires in national forests and parks annually affect millions of acres of
public lands, destroying recreational sites, historical areas, and scenic wilderness, and costing
taxpayers hundreds of millions of dollars every year in suppression costs and lost resources. This
research examined the spatial correlation between human activities and human caused wildfire
occurrences within the Willamette National Forest to explore whether these activities might be
responsible for many wildfire ignitions. Between 1995 to 2008, 493 human caused fires
occurred. The density of these fires was investigated to identify clustering near recreational sites
and human infrastructures. Maxent was used to model the probability of wildfire occurrences in
relation to the recreational sites and human infrastructure areas located throughout the Forest.
It was initially hypothesized that more wildfires occur near specific kinds of recreational
sites than elsewhere. Preliminary data exploration showed high densities of wildfire occurrences
near the towns, human infrastructures, and major highways although these were also areas of
clusters of recreational sites. Thus, it was not possible to identify visually which particular
activities were most strongly related to wildfire ignitions. Maxent results revealed that areas of
high population densities and recreation site clusters were more likely to correspond to areas of
more human caused wildfire ignitions.

1

Chapter 1 Introduction
Since 1944, a lone bear has been reminding people to be careful about forest fires. Today after
70 years of propaganda, wildfires are still a threat to property, lives, wildlife, and resources.
Carelessness, accidents, and arson account for a vast majority of human caused wildfires (Chas-
Amil et al. 2015; Fox et al. 2015; Pew 2001). These events cost hundreds of millions of dollars
in man hours and equipment to suppress and in lost revenue for resource companies and tourism
organizations. Research has shown that when rates of human activity within national forests and
parks increase, the opportunity for wildfire ignitions that are associated with those activities will
rise (Manning 2001). Thus, it is imperative to understand that human actions and wildfires are
related.
This research investigates the spatial relationship between human activity and wildfire
ignitions to assess the human caused ignition risk of wildfires within the Willamette National
Forest (WNF) in Oregon. Figure 1 shows the location of the main settlements and roads in and
around it. Given that the WNF is within a few hours’ drive from six of Oregon’s larger cities—
Salem, Albany, Corvallis, Eugene, Redmond, and Bend—the WNF attracts a large number of
visitors who partake in recreational activities available throughout the Forest. Industrial activities
such as logging are common throughout the WNF and traffic from five major highways are also
potential human causes of wildfire ignitions. With the abundant spatial data available from the
Forest Service and Data.Gov, the WNF is an excellent case study area with which to explore the
spatial relationship between human activity and wildfires.
2

Figure 1. The Study Area: The Willamette National Forest, major roads, towns and communities,
and nearby major cities.
Lightning caused wildfires were removed from the study in order to focus on human
activities and their relationships with wildfires that may occur due to those activities. Though
lightning is a major cause of wildfires, environmental factors (e.g. humidity and fuel load) that
may lead to wildfires were not considered for this study, as it was hoped that by investigating the
sources of the human caused wildfires, a better assessment of human caused wildfire ignition
risk would be reached. Figure 2 shows the human caused wildfire ignition distribution in relation
to lightning caused wildfire ignition locations. The two distributions of wildfire ignitions are
sufficiently different between lightning and human causes to see there is no relationship between
3

the two. This suggests a different mechanism driving the human caused ignitions and supports
the decision to focus only on those fires and the location of human activities in this study.

Figure 2. (A) Locations of human caused wildfire ignitions, (B) locations of lightning caused
wildfire ignitions, and (C) combined human and lightning caused wildfire ignitions.
1.1. Wildfires and the Willamette National Forest
Like many national forests across the United States, the Willamette National Forest has
often dealt with the adverse effects of wildfires. Throughout time, wildfires of varying sizes have
altered the landscape and changed the way humans utilize the Forest.
1.1.1. History of Wildfires in the Willamette National Forest
The original Native American settlers of the Willamette Valley and the Western Cascades
used fire to maintain harvesting areas and hunting grounds throughout the entire Willamette
4

Valley and surrounding foothills (Burke 1979; LeQuire 2010). The wildfires suppressed
undergrowth that allowed fire-hearty trees and shrubs to thrive. As Burke (1979) has pointed out,
the Native American method of using fire as a form of forest management helped to reduce any
naturally occurring wildfires.
The issue of human caused wildfires within the Western Cascades actually began in the
late 1800s when sheep farmers were blamed for starting fires to increase their grazing areas
(Rakestraw 1991). Combating the idea of purposely starting a fire on public domain lands was
addressed in 1896 with The Act of February 24, 1897 (29 Stat., 594) which
provided penalties for willfully or maliciously setting on fire any timber, underbrush, or
grass on the public domain; carelessly or negligently leaving fire to burn unattended near
any timber or other inflammable material; or failing to totally extinguish any campfire or
other fire in or near any forest, timber, or other inflammable material before leaving it
(Rakestraw 1991, 21-22).
Then in 1897, The Sundry Civil Appropriations Act of June 4 (30 Stat. 11, 34) granted the
Secretary of the Interior the authorization to make rules that would protect the Forest Reserves
against destruction by fire (Rakestraw 1991). Various additional laws since that time have been
imposed to try to improve the safety and protection of the forests, but wildfires persist. During
the 13-year time span the data for this study cover, a total of 1,584 wildfires, both natural and
human caused, were recorded within the WNF. Data from the Willamette National Forest lists
nine different wildfire causes which include lightning, equipment use, smoking, campfires,
debris burning, railroad, arson, children, and miscellaneous.
1.2. Common Activities Found in the Willamette National Forest
Every year the Willamette National Forest receives numerous visitors seeking some type
of recreation or engaging in some sort of occupation. There are many different types of
5

recreation and industrial activities within the WNF. All of these activities are a necessary part of
what creates the uniqueness of the WNF.
1.2.1. Recreational Activities in the Willamette National Forest
Recreational areas within the Willamette National Forest encompass a large variety of
activities. Camping, hiking, fishing, boating, hunting, skiing, horseback riding, and off-road
vehicle use are some of the most common activities in which visitors engage. Trails of varying
difficulty and length allow people to explore the multiple attractions the WNF has to offer. Some
trails, such as the Pacific Crest Trail, test hiker’s endurance with a variety of difficulties and long
travel times. Hunting and fishing bring visitors to the WNF, allowing them a chance to hunt a
variety of different game animals or fish the multiple lakes and rivers for several different types
of trout, steelhead, and salmon. Camping allows the visitors a chance to relax and enjoy the
outdoors. Some trails and gravel roads are accessible to off-road vehicles, such as motorcycles,
ATVs (all-terrain vehicles), and other specialized vehicles. Horseback riding is allowed on
certain trails, and some campsites are designed to accommodate horses and their owners. All
Forest Service maintained campgrounds and day use areas have some type of fire pit, fireplace,
or barbeque station (U.S. Forest Service n.d.).
1.2.2. Industrial Activities in the Willamette National Forest
The Willamette National Forest is no stranger to industrial activities. Industrial activities
found throughout the WNF include logging, construction, hydroelectric dams, high voltage
power line and cell phone tower repairs, fish hatcheries, trail repair, and railroads. Each quarter,
the Willamette National Forest releases a Schedule of Proposed Actions (SOPA) that details
projects planned for the WNF. These projects range from repairing trails and fixing winter road
damage to upgrading campsites and rehabilitating huckleberry habitats. Manmade structures
6

such as hydroelectric dams, cell phone towers, and ski lifts often need to be upgraded or repaired
due to normal wear and tear. These activities often require a significant amount of manpower and
can take weeks, even months to complete. The U.S. Forest Service and the Oregon Department
of Fish and Wildlife each maintain a fish hatchery within the Willamette National Forest. Each
of these fish hatcheries has abundant human and vehicle traffic entering and exiting the facility
due to the activities of volunteers, employees, visitors, and fish hauling tankers.
Some less common harvesting and collecting activities can be classified as either
recreational or industrial depending on whether resources are collected for income. In Willamette
National Forest, these activities include collecting moss, harvesting camas or wild berries,
gathering mushrooms or truffles, or picking wildflowers or other types of decorative foliage.
Such activities are usually undertaken by single individuals or small groups who have acquired
the proper permits and have been given suggestions about the location of appropriate collection
areas within the WNF.
1.3. Wildfire Impacts on Economic Factors
Historically, wildfires have been described as both a tool for and a disaster to human
communities (Rakestraw 1991). Wildfire before the European settlers was a natural cycle of the
forest. Native Americans would follow a wildfire after a few days and harvest seeds and insects.
European settlers then used fire to clear land for harvesting and to remove remaining stubble
after harvest. It was not until the great forests of the New World were finally deemed a resource
instead of a nuisance that wildfires became an issue.
Wildfire suppression costs have increased dramatically due to human infrastructure
within the forest (Donovan 2008; Pew 2001; Gebert 2007). The National Interagency Fire Center
releases details on the cost of fighting wildfires every year. Table 1 displays the total firefighting
7

costs for the entire United States for the years 1995 and 2008, the first and final years examined
in this study. Note that although the number of fires declined, spending and acreage affected
increased. Costs of manpower, small equipment (personal equipment, chain saws, hoes, etc.),
heavy equipment (bulldozers, fire trucks), air support, and logistics make up the majority of the
necessary expenditures for every wildfire (National Interagency Fire Center 2016). Furthermore,
the table does not show the costs of lost resources, homes, wildlife, domestic animals, revenue,
and scenic beauty.
Table 1. National Interagency Fire Center costs for the years 1995 and 2008 for wildfire
suppression. Source: National Interagency Fire Center 2016.
Year # of Fires # of Acres
Dollars spent (in million $)
Forest Service DOI Agencies Total
1995 82,591 2,896,147 161.5 78.4 239.9
2008 78,979 5,292,468 1,193.1 392.8 1,585.9

Wildfire damage reduces the availability of resources such as paper, pulp, and lumber to
wood industries (Flannigan, Stocks, and Wotton 2000). Damage to hiking trails, campgrounds,
and other recreational sites or surrounding areas prevent visitors from entering the areas until
they have time to recover. Income can be lost from fees associated with permits, dues, and
programs that are dependent on the availability of activities in or around recreational areas.
Wildfires also impact local businesses: loss of tourism income due to visitors avoiding the effects
of the fires or unable to enter fire-affected areas also impact local economies.
1.4. Project Objectives
Wildfires caused by human actions are a serious threat to private property, national parks,
and forests. The goal of this study was to examine the relationship between the location of
human activities and the location of wildfire ignitions in order to gain some insight about the
potential locations of future wildfires. This study focused specifically on recreational sites and
8

other human infrastructures and their relationship with human caused wildfire ignitions within
the Willamette National Forest. The main objectives of this study were to validate three
hypotheses:
(1) campfire caused fires would cluster near campgrounds,
(2) smoking related fire ignitions would be more abundant near day use areas, towns,
and residential areas, and
(3) wildfires would be more frequent near areas with multiple numbers of recreational
sites or human infrastructures.
1.5. Organization of this Document
This remainder of this thesis is composed of five chapters. Chapter 2 reviews related
work about wildfires and their relationships to human activities and human infrastructures. This
chapter also introduces previous research on the modeling of wildfire probability. Chapter 3
discusses the data pertinent to the study as well as the preliminary data exploration that was
completed. Chapter 4 explains the Maxent software used to model the relationship between
activities and wildfire ignitions. It also explains how the data were converted for use in Maxent
and the necessary processes for entering the data within the modeling software. Chapter 5
provides the results of the Maxent software modeling processes and discusses the probability of
occurrence of wildfires given different variables related to human infrastructure or recreational
activities within the WNF. Chapter 6 offers discussion about the findings of this thesis and
possible future work.

9

Chapter 2 Related Research
A great deal of research has gone into investigating the causes of wildfires. Previous studies have
shown that human activities in areas like the WNF are often responsible for wildfire ignitions
(Pew 2001). This chapter takes a look at prior research that explored factors that influence
wildfire occurrences, including recreational activities, human infrastructure, industrial activities,
forest management, and weather conditions. It also introduces Maxent, a machine learning tool,
that has been used to model relationships between various environmental variables and species
locations using presence only data.
There has been extensive research on how human infrastructure encroachment on
wilderness areas affects wildfire probability (Chas-Amil et al. 2015; McLemore 2017). Much of
the previous work deals with the Wildland Urban Interface (WUI). Wildfire occurrences that are
spatially related to recreational activities within the WNF are the focus of this study, so it is
important to understand that WUI deals strictly with areas of urbanization and some, but not all,
forms of human infrastructure. Urbanization is associated with growth or expansion of cities or
areas of large population densities, whereas human infrastructure for the purpose of this study,
refers to those areas related to highways, roads, railroads, resorts, private residency, towns or
communities, and administrative areas.
2.1. Human Activities Related to Wildfire Occurrences
Forests provide numerous opportunities for humans to enjoy and use the resources within
them. Some activities are industrial, such as logging, road building, and forest management.
Other reasons for people to trek into the WNF are recreational, including activities such as
camping, hiking, fishing, cross-country skiing, or off-road vehicle usage. According to Parisien
et al. (2012), human activities need to be included in all estimates of wildfire probabilities
10

because human activities as potential factors in wildfire occurrences are often ignored. Whatever
the reason for humans to be in the WNF, there is always a possibility of wildfires.
2.1.1. Relationship of Recreational Activities to Wildfire Occurrences
With a vast number of recreational opportunities, the Willamette National Forest receives
a considerable number of visitors each year. Annually, the number of visitors to national forests
continues to increase (U.S. Forest Service 2014). The results of several studies suggest that as the
popularity of outdoor recreation grows, and the rate of yearly visitors increases, so does the risk
of wildfire ignitions from human related activities. Camp and Krawchuk (2017) showed that
recreational activities are positively associated with wildfire ignitions. Other research has shown
that areas associated with activities such as camping and hiking, particularly those involving
campfires, create situations that increase the possibility of wildfire ignitions (Chas-Amil et al.
2015; Cole 1981). Romero-Calcerrada (2008) showed that distance from recreational areas is an
important factor when predicting wildfire ignitions. Pew (2001) discovered that recreational
activities on Vancouver Island in Canada contributed to a high percentage of wildfire ignitions,
though most of the fires were small.
Recreational destinations such as the Willamette National Forest allow human
recreational activities into back country areas where fire suppression response may be inhibited
due to inaccessible terrain. These types of situations, where a wildfire ignition happens in remote
areas, may increase the acres affected due to the time it takes to report the wildfire. Hiking and
horseback riding are examples of activities that allows people access to areas of the WNF where
a wildfire may go unnoticed. Camp and Krawchuk (2017) note that campfires that are built while
hiking pose a high risk of wildfire probability. Romero-Calcerrada (2008) showed that trails used
11

for hiking or horseback riding increase the risk of wildfires by increasing accessibility to areas
inside forests.
Additionally, areas with high levels of hiking and horseback riding are subjected to
changes in vegetation composition, which allows non-native plant species to invade the areas
(Cole 1981). According to Stein et al. (2013), replacement of native vegetation by highly
flammable non-native vegetation increases the potential for wildfires.
2.1.2. Relationship of Human Infrastructure to Wildfire Ignitions
There has been extensive research on how human infrastructure encroachment on
wilderness areas affect wildfire probability by increasing the frequency of human activities in
surrounding forested areas (Chas-Amil et al. 2015; McLemore 2017). Human infrastructure in
this research was defined earlier as highways, roads, railroads, resorts, private residences, towns
or communities, and administrative areas.
Camp and Krawchuk (2017) have shown that greater densities of infrastructure increases
human activities in forested areas. Such human infrastructure can be found throughout the
Willamette National Forest, except where the area is designated as a Wilderness Area. The next
sections discuss the impact of different kinds of human infrastructure individually.
2.1.2.1. Roads
Originally, roads were built to gain access to those resources (timber, ore, coal) that a
growing nation required. Today, roads into the forest have multiple uses such as hunting,
logging, and forest management and they are used by a variety of people. Yang et al. (2007) have
shown that human caused wildfire occurrences are more likely to be spatially clustered near
human infrastructures such as roads and agricultural areas. Findings have also shown that the
rate of wildfire occurrence per unit area is higher near roads and decreases with distance from the
12

road (Pew 2001; Romero-Calcerrada 2008). This correlates with a study by Young (2015) that
showed higher frequency of wildfire occurrences along major freeways across the Western
United States. Roads, according to Camp and Krawchuk (2017), contribute to wildfire ignitions
through discarded cigarettes, sparks from accidents, or even a bad muffler.
2.1.2.2. The Wildland Urban Interface and Wildfires
The Wildland Urban Interface (WUI), as defined by Stein et al. (2013), comprises areas
in or around forests, grasslands, shrub lands, and other natural areas into which some type of
urban expansion is occurring. In a report for the U.S. Forest Service, Stein et al. state that
approximately 32 percent of U.S. housing units (homes, apartment buildings, and other human
dwellings) are within the WUI. Stein et al. also note that as many as one million homes were
built within the WUI in Oregon, California, and Washington between 1990 and 2000. For the
purpose of this study in the Willamette National Forest, the WUI includes all areas surrounding
towns, communities, areas of summer homes, lodges, resorts, and hotels.
Population expansion into forests and rural areas has been shown to be related to an
increase in wildfire ignition counts by as much as double the number of natural fires (Calef,
McGuire, and Chapin 2008). Proximity to population centers was also associated with higher
counts of arson related wildfires (Yang 2007). Additionally, Hawbaker et al. (2013) discovered a
positive relationship between urban density and increases in wildfires in the Mediterranean
California climate zone.
2.1.3. Relationship of Industrial Activities to Wildfire Occurrences
There are many reasons for humans to enter the forest, but one of the oldest reasons
would be to collect resources. Timber harvesting practices and replanting methods create denser
stands of smaller tree sizes allowing for fire to crown (reach the upper levels of the trees) and
13

spread faster than in a forest that had seen fires previously (Stephens et al. 2009; Lenart 2006;
Pew 2001). With advances in logging techniques, it was anticipated that as a bonus, logging
could reduce potential wildfire occurrences and severity by reducing fuel loads. Yet, research has
shown that logging causes reoccurrence and greater severity of wildfires due to an unintended
effect of increasing fuel loads (Camp and Krawchuk 2017; Lenart 2006; Pew 2001; Westerling et
al. 2006). McKenzie et al. (2004) explain that increasing temperatures within a region due to
current logging practices will speed up the drying of fuel loads which will increase the severity
and frequency of wildfires.
In a study done on Vancouver Island, Pew (2001) discovered that logging between 1950-
1992 was responsible for the greatest proportion of the increase in large acreage fires. In a study
done by Miller et al. (2009) on fire severity in the Sierra Nevada Mountains, they found that fire
severity increased overall across their period of research. Their data showed that increases in fire
severity was due to fragmentation of forests (the patchwork like results of logging sections of a
forest and leaving other sections either uncut or in different phases of growth) (Miller, Safford,
and Crimmins 2009).
Wildfire occurrences have been attributed to timber harvesting which leaves unwanted
material (commonly referred to as slash) that dries and increases potential ignitions (Camp and
Krawchuk 2017). Pew (2001) confirmed that slash burns, the burning of unwanted material after
harvesting operations, were directly related to 29 percent of all logging related fires on
Vancouver Island in British Columbia, Canada.
2.2. Wildfire Prevention Through Forest Management
Maintaining a healthy forest ecosystem is imperative to reducing wildfire potential and
severity, aiding suppression, and insuring safety of human visitors in case of wildfires. After the
14

devastating wildfires of 1910, where 3 million acres were destroyed, and 78 firefighters lost their
lives, state and federal agencies made it a high priority to suppress all wildfires (Stein et al.
2013). Later in the early 20
th
century, wildfire occurrences were lessened by different types of
forest management, including techniques such as grazing that were used to reduce grasses, small
shrubbery, and leaf litter thereby limiting fuels on which a fire could burn (Lenart 2006;
Westerling et al. 2006). During the 1960’s and 1970’s, the importance of wildfires’ role in
forests and wildland ecosystems was beginning to be understood (Stein et al. 2013). Since that
time, controlled burns have been carefully reintroduced to restore fire-adapted vegetation and
reduce fuel loads.
As of 2009, within the Western U.S. over ten million hectares of forests are considered a
medium to high risk for wildfires and the potential for wildfire ignitions is a significant issue for
forest management (Stephens et al. 2009). According to Lenart (2006), one of main goals of
current forest management practices is to reduce wildfire dangers. Hirsch et al. (2001) explain
that “Fire-Smart Forest Management” uses methods such as thinning existing stands of timber,
harvest scheduling, and harvest designs to reduce potential wildfire severity.
2.3. The Role of Weather in Wildfire Occurrence
Weather in the Willamette Valley and the neighboring foothills and mountains is ever-
changing. There are many contributing factors to the sometimes-erratic nature of the weather.
Terrain, vegetation, ocean currents, and even inland waterbodies influence different attributes of
weather. Terrain is influential in forcing winds and clouds, vegetation can increase humidity,
lakes and streams create areas of cooler temperatures, and ocean currents contribute to all types
of weather, especially yearly weather patterns. Weather changes due to yearly cycles of El Niño
and La Niña create abnormalities such as drought, unusual precipitation patterns, early spring
15

temperatures, longer wildfire seasons, and increased wildfire severity (Dale et al. 2000;
Westerling et al. 2006). According to Stein et al. (2013), forests and other wildland areas with
low humidity and higher seasonal temperature patterns have a higher susceptibility to wildfires.
2.3.1. Fair Weather Activities and Impact on Fire Frequency
Wildfires in the western United States are a seasonal occurrence, with 94 percent of
wildfires occurring between May and October (Westerling et al. 2003). Seasonally, the Cascade
Mountain Range has the highest number of wildfires between July and August. Recreational
activities are often most prominent during the warmer seasons of the year, which coincides with
fire season. It is this combination of warm, dry weather which brings people to the forests, but it
is also a combination for wildfires. Romero-Calcerrada (2008) also found that wildfire
occurrences increased near recreational areas during the summer holiday months, July and
August. Pew (2001) observed that monthly percentages of the annual count of human caused
wildfires on Vancouver Island began to increase in May, peaked in August, and then dropped
substantially in October. According to Young (2014), peaks in wildfire occurrences were
recorded during certain times of the year, including the beginning of hunting season in the late
Summer or early Fall, the Fourth of July, Memorial Day, and Veteran’s Day.
Weather often determines the intensity of human activities in the WNF. With warmer
weather, people look forward to hiking their favorite trails, camping, or undertaking other
recreational activities and may make plans to engage in their favorite activity as early as
possible. For example, Monahan (2016) concluded that warmer than normal early spring
temperatures increases the number of visitors and extends the visitation seasons within National
Parks.
16

2.4. Modeling Wildfires using Maximum Entropy Techniques
Maximum entropy is a machine learning technique that is used to model presence only
data against environmental conditions to predict the probability of presence for the purpose of
species distribution and environmental niche modeling (Merow 2013). This method has been
used extensively to model the relationship between the environment and the occurrence of
animals, plants, and fish, as well as archaeological sites and other kinds of sites (Elith et al. 2011;
Oyarzun 2016).
There has been some research on the use of maximum entropy to model wildfire
probability. Wildfire ignition distribution modeling is similar to species distribution modeling
because the basic approach is to analyze locations in relation to environmental variables said to
influence the spatial distributions of wildfire ignitions.
According to Massada et al. (2013), human and natural wildfires display distinctive
spatial patterns that can be determined by using human and environmental variables. Using 340
human caused ignition points and three types of environmental data (landcover, topography and
human), Massada et al. created three different models of wildfire ignition probabilities within the
Huron-Manistee National Forest. They found that in areas where human population density is
high, and the number of human caused wildfires is greater than the number of naturally occurring
wildfires, accessibility and density contribute more to the wildfire probability. Areas near roads
also showed higher probabilities of human caused wildfires than naturally occurring wildfires.
In a study conducted by Parisien et al. (2009), they attempted to quantify the influence of
environmental variables on the prediction of wildfire distributions for three separate spatial
scales: the conterminous United States, the state of California, and five fire-prone ecoregions of
California. To model their hypothesis, they used both Maxent and boosted regression trees
17

(BRT) and compared the models. Fire data for their research dated from 1950-2003 and omitted
prescribed burns and wildfires that measured under 300 acres. For their environmental variables,
Parisien et al. included climate data, elevation, vegetation, and areas with sparse vegetation. To
compare the models, they evaluated the receiver operating characteristics (ROC) and the area
under the curve (AUC) for the highest score. The higher the AUC score is, the better the model
fits real world applications. The results indicated that Maxent and the BRT showed similar
results across all models. The California model results produced the highest AUC score across
all models, and the five ecoregions had mixed results, with scores varying between high desert
and coastline regions.
The use of machine learning software to model wildfire probabilities for evaluating
relationships of fire presence against environmental variable backgrounds has been shown to
produce probability maps that capture complex relationships (Parisien et al. 2012). Massada et al.
(2013) question the fact that comparing continuous probability maps is not a simple task and
suggests comparing both values and patterns. Maxent has been used to predict wildfire
probabilities in both large areas and in smaller, localized areas. In both situations, research has
shown that Maxent is capable of modeling wildfire ignition points with relative accuracy.
2.5. Summary
Understanding that there are many different factors that contribute to wildfires is
important. The previous research discussed in this chapter provides information about how
different activities, practices, and natural conditions affect how wildfires begin and how they
react to those factors. In all, the main cause that was referred to the most was some sort of human
activity.
18

Chapter 3 Data and Preliminary Data Exploration
This chapter provides a description of the study area, gives an overview of the data sources, and
explains the preliminary data exploration used in this study. The purpose of the preliminary data
exploration was to examine the data for any spatial correlation between wildfire occurrences and
recreational sites, administrative locations, and/or human infrastructure. The results of the
preliminary analysis indicated the direction to be taken in the full analysis described in
Chapter 4.
3.1. Study Area: Willamette National Forest
The Willamette National Forest is the area of focus for this study. The WNF is the fourth
largest national forest in the state of Oregon (U.S. Forest Service n.d.). The WNF is centrally
located along the western slopes and foothills of the Cascade Mountain Range bordering the
eastern side of the Willamette Valley. With four major cities to the west of the WNF and two
major cities to the east of the WNF, traffic and visitors are abundant, as are recreational
activities.
3.1.1. Geographical Features of The Willamette National Forest
The Willamette National Forest is 110 miles long and is spread across six Oregon
counties (U.S. Forest Service n.d.). Approximately 1,675,400 acres are designated as National
Forest under the supervision of the U.S. Forest Service. Under the Wilderness Act of 1964,
380,805 acres of the WNF were designated as wilderness area, which forbids human residency,
motorized vehicles, and requires no disruption of natural ecology. Figure 3 shows the main roads
and communities in and surrounding the WNF.
19

Figure 3. The Willamette National Forest with town, community, highway locations, and
privately owned areas outside of the WNF.
Four towns and two unincorporated communities are located within the boundaries of the
Willamette National Forest. The towns of Detroit and Idanha are located near the northern
20

boundary along Highway 22. The unincorporated towns of McKenzie Bridge and Rainbow are
found near the western boundary on Highway 20. The towns of Oakridge and Westfir are located
along Highway 58 which runs through the southern quarter of the WNF. The towns of Detroit,
Idanha, and Westfir have populations of less than 260, and Oakridge maintains a population of
3205 (U.S. Census Bureau 2010). Population counts for Rainbow and McKenzie Bridge were not
available. All towns and communities inside the boundary of the WNF, while containing a
combination of privately owned lands and areas governed by local, county and state agencies, are
considered part of the WNF for this analysis.
An additional 123,000 acres of small inholdings are scattered throughout the WNF.
These areas fall under the supervision of private ownership or other government agencies but are
considered part of the WNF for the purposes of this study. Some of these privately held lands are
used for forestry purposes only. The areas were considered part of the WNF for this study based
on the fact that often there is no marked boundary or other evidence to differentiate the WNF
from private lands. Another factor in the decision was that U.S. Forest Service roads pass
through private lands, making these privately held lands accessible to the public. Other types of
privately held lands are discussed later in this chapter.
A few small, isolated areas of the WNF fall outside of the main forest boundary; these
small islands of national forest are visible on the west side in Figure 3. While these islands are
part of the national forest, they do not contain any form of recreational site or human
infrastructure, nor do any wildfire ignition points fall within them. These islands were cut from
analysis, but they are displayed in maps presented in this thesis.
As well, there are some larger privately-owned areas that are holes inside the main forest
boundary near the western mid-section of the WNF (shown in blue in Figure 3). These areas
21

were removed from the data set used in this study by the U.S. Forest Service prior to distribution.
As a result, these areas are not considered part of the privately-owned areas that were included in
the analysis.
A variety of terrains are found within the WNF, from narrow canyons, cascading streams,
and wooded foothills to a tall stratovolcano at the eastern forest boundary (U.S. Forest Service
n.d.). The WNF elevation ranges from approximately 900 feet above sea level to 10,495 feet atop
Mt. Jefferson and the Three Sisters.
3.1.2. Recreation in the Willamette National Forest
Recreational areas within the Willamette National Forest encompass a large variety of
activities. Camping, hiking, fishing, boating, hunting, skiing, horseback riding, and off-road
vehicle usage are some of the most common activities visitors engage in. Within the WNF there
are 90 campgrounds (82 Forest Service, 8 private or other agency), and 22 picnic or day-use
areas. All Forest Service maintained campgrounds and day use areas have some type of fire pit,
fireplace, or barbeque station (U.S. Forest Service n.d.). There are 1360 miles of hiking trails
give access to a large amount of the WNF (U. S. Forest Service n.d. A). There are 400 named
lakes, 4101 secondary waterbodies equaling approximately 26,000 acres of surface water, and
2700 miles of rivers and streams for boating, fishing, and rafting (U.S. Forest Service n.d.).
There are two developed ski areas, Hoodoo Ski Area and Willamette Ski Area, both operating
under special permits within the Willamette National Forest. There are certain areas within the
WNF that allow for ATV recreation where ATVs can be used on specific trails with the proper
permits.
Wildfire safety warnings and information are common propaganda throughout designated
recreational areas. Despite these warnings, wildfires do occur.
22

3.1.3. Vegetation
The WNF is positioned along the Western Cascade foothills and contains some of the
most productive forest land within the U.S (U.S. Forest Service n.d.). The forest vegetation
consists mainly of Douglas Fir, but also contains 15 different conifer species and other broad-
leaf hard and softwoods. Old-growth stands of Douglas Fir are common throughout the WNF
with tree diameters ranging from 3-8 feet (U.S. Forest Service n.d.). Douglas Fir and other types
of timber are an important resource for local economies that depend on the lumber and wood
fiber industry. Other forest resources such as moss, mushrooms, camas, and Christmas trees are
also commonly sought-after commodities within the WNF by Native American tribespeople and
others in small groups or individually.
The Willamette National Forest is more than just a stand of timber. Throughout the WNF
there are meadows of grasses, shrubbery (underbrush), a wide variety of wildflowers, different
types of berry plants, and lichens. All of the different plant life makes up a well-established
ecosystem that depends on seasonal climate fluctuations.
3.1.4. Climate
The Willamette National Forest is located at a geographic location and latitude that gives
it the classification of a sub-tropical rainforest. This means it receives between 40 to 150 inches
of rainfall, some of which falls as snow in the upper elevations (U.S. Forest Service n.d.). With
its location against the western side of Cascade Mountain Range, and being approximately 85
miles inland from the Pacific Ocean, a number of different climates affect the WNF. Depending
on location, elevation, and season, the temperatures of the WNF can vary quite dramatically,
with temperatures ranging from single digit winter temperatures to summer temperatures into the
mid 80’s Fahrenheit.
23

Cyclical climate patterns such as El Niño and La Niña are common occurrences that
affect the climate within the WNF. Depending on the cycle, a variety of weather patterns from
drought to intense rain/snow storms are possible. Droughts brought on by the El Niño cycle
reduce precipitation and available snowpack in the mountains. La Niña denotes a cycle where
precipitation levels are higher than normal, which helps to increase snowpack in the higher
elevations and to maintain lower temperatures during the spring and summer seasons.
3.1.5. The Role of the Forest Service and Other Agencies
The Willamette National Forest falls under the federal authority of the U.S. Department
of Agriculture (USDA). The USDA manages national forests throughout the United States
through the U.S. Forest Service (USFS). The USFS in its mission as managers of the Willamette
National Forest has laid two principles as foundations for maintaining the WNF
-Land Stewardship: The Willamette National Forest is managed to conserve natural
resources, promote long-term productivity and sustained yield, and enhance
environmental quality. Commitment to long-term stewardship must be demonstrated by
strong and visible sensitivity to the land in on-the-ground management activities. (U.S.
Forest Service 1990, IV-2)
-Public Trust: Managers of the Willamette National Forest are public servants and are
charged to listen and to provide for the public needs to the best of their ability. They will
be open and forthright with the public in all manners. Regardless of the potential conflict
and controversy, public interest in National Forest management is best served by active
and informed public participation. (U.S. Forest Service 1990, IV-2)
The USFS oversees an assortment of resource management programs that pertain to the
WNF. Wildfire management, hydrological concerns, invasive plants, timber management, and
recreational management are a few issues that are discussed on the Willamette National Forests
Project page https://www.fs.usda.gov/projects/willamette/landmanagement/projects. The USFS
is charged with maintaining all federally owned recreational sites within the WNF: this means
upkeep of all camping sites, boating areas, trails, roads, and permits for ski areas. Hydrological
24

or aquatic programs target the rehabilitation of streams, rivers, and neighboring banks and
vegetation for the improvement of aquatic species. Wildfire management within the WNF
consists of assessment of wilderness areas, monitoring vegetation and timber, as well as
understanding and modeling fundamental fire processes, interactions of fire with
ecosystems and the environment, social and economic aspects of fire, evaluating
integrated management strategies and disturbance interactions… [and] prioritizing and
implementing fuel hazard reduction projects, in smoke forecasting, in rehabilitating and
restoring land after severe wildfire, and in providing information to home owners in the
wildland-urban interface (U.S. Forest Service 2016b, n.p.).
The State of Oregon maintains the five major highways that run through the WNF and
polices the highways and urban areas found within the WNF. The State of Oregon operates
maintenance yards near the highways to accommodate highway maintenance year-round. The
Oregon Department of Fish and Wildlife is responsible for overseeing all hunting and fishing
regulations, licensing, monitoring wildlife and aquatic species populations, as well as
enforcement of hunting and fishing rules and regulations within the WNF.
3.2. Data Overview
The main objective of this study was to determine if there exists a spatial relationship
between wildfire ignitions and recreational sites and human infrastructures in the WNF. This
study focused solely on wildfires that have been determined to have been started by human
related activity, removing natural occurring fire ignitions from the equation.
Data were collected from one or all of the following data libraries: the Willamette
National Forest Data Library, USDA Forest Service FSGeodata Clearinghouse, and Data.gov.
All data for this study was converted to NAD 1983 USFS R6 Albers, a version of the Albers
Conical Equal Area system. This is the coordinate system the Forest Service uses in Region 6 of
the USDA. Each data type and how it was used is explained in further detail in the next sections.
25

3.2.1. Forest Boundary Data
The Willamette National Forest boundary data is a polygon layer that defines the forest
boundary and other ownerships found within the forest boundary, as well as depicts the
individual ranger districts that make up the complete national forest. The Willamette National
Forest Headquarters in Eugene, OR created the data for the purpose of management of natural
resources. The data attributes used in this study are the DIST (ranger district) and
PROCLAIMNF (this indicates whether polygons contain land inside or outside of the WNF, as
explained earlier in the study area description). This data set was created February 2, 2001 and
was last updated September 13, 2007. Although this boundary was more than 10 years old at the
time of this study, the wildfire data covers the years up to 2008 so it is current with respect to the
other data.
3.2.2. Wildfire Data
The USFS recorded 493 human caused fires between 1995-2008 in the WNF. The time
range for this study was chosen based upon the time span of the original data found within the
Willamette National Forest Data Library. The wildfire data set from the Willamette National
Forest Data Library was incomplete since data between the years 1999 to 2003 were missing;
thus, that data set was discarded. It was necessary to use data found in the FSGeodata
Clearinghouse and the Data.gov Fire Program Analysis - Fire Occurrence Database which
included data from all 50 states, Guam, and Puerto Rico. Figure 4 displays the locations of the all
human caused wildfire ignitions within the Willamette National Forest between the years 1995 to
2008, with many overlapping symbols where fire ignition points are superimposed. Table 2
summarizes the wildfires during 1995-2008 in the WNF by category.
26

Figure 4. Location of the 493 human caused wildfires in the WNF between 1995 and 2008.
27

Table 2. Number of human caused wildfire occurrences by category, 1995-2008
Wildfire Type
Number of
Occurrences
Arson 34
Campfire 256
Children 13
Debris Burning 17
Equipment Use 20
Smoking 50
Miscellaneous 93
Railroad 10
Grand Total 493

The wildfire data selected for this study were obtained in point format. A point format
was chosen because the idea of this study was to find a relationship between the original point of
ignition to whatever human activity might be responsible for the cause. The USFS collects and
stores fire occurrence data in a database called the Fire Statistical System (FIRESTAT) (U.S.
Forest Service 2016a). The data for each individual wildfire occurrence is recorded using a form
called the Individual Wildland Fire Report. Current methods utilize GPS technology to record
the wildfire ignition locations, but prior to 2014, the locations were input using either lat/long
coordinates taken at the occurrence site (small fires) or using a map to determine the location of
the fire ignition using the Township, Range, Subsection, and Principal Meridian of that location
to identify and plot the coordinates.
The U.S. Public Lands Survey System uses Township, Range, Subsection, and Principal
Meridians to subdivide the United States. Townships are measured as 6-mile by 6-mile squares.
Townships are surveyed north, south, east, and west from an initial point. There are 37 Principal
Meridians running parallel, north to south across the United States. A baseline runs
perpendicular to the meridians and is used to designate Township locations north or south of the
28

baseline. A range number indicates the number of townships east or west of the Principal
Meridian. Townships are further subdivided into 36 Sections, which are a 1-mile by 1-mile
square. These Sections, too, can be subdivided into Subsections which are measured ¼-mile by
¼-mile and contain 40 acres. Assuming the center points of Subsections were used to get the
coordinates of fire ignition locations, the precision of these locations recorded in this database
should be considered +/- 0.125 mile, 660 feet or approximately 200 m.
The original data were filtered to extract only data found in the Willamette National
Forest. Further editing included clipping the data to remove any outliers outside of the forest
boundary as described in the previous section and selecting only the data from 1995-2008. All
non-human caused wildfires were removed. The attributes used for this study were the
FIRE_NAME, STAT_CAUSE (a value given to identify the wildfire cause: for example,
Campfire = 4), STAT_CAUSE_1 (a written description of the wildfire cause), X_COORD,
Y_COORD. Eight different categories are found in the STAT_CAUSE (wildfire cause) attribute
data: Arson, Campfire, Children, Debris Burning, Equipment Use, Smoking, Railroad, and
Miscellaneous.
Only the wildfire causes of Smoking (50 fires) and Campfires (256 fires) were deemed to
have sufficient occurrences to permit detailed individual examination of these causes. These data
were extracted into two separate data layers for individual analysis. Data in the Miscellaneous
causes category (93 fires) while numerous, were not examined as a separate category due to
uncertainty of what “miscellaneous” was categorizing—did these include fires in other
categories? Finally, the full set of all 493 wildfires were included in analyses that considered the
entire collection of fires collectively. Maps showing the locations of each type of wildfire
ignition are included in Appendix A.
29

3.2.3. Recreational Site and Human Infrastructures Data
Creation of the recreational site and human infrastructures data layers required the
combination of data from several sources. These are shown in Table 3 and discussed below.
Table 3. Sources used to create the recreational site and human infrastructures data layers
Layer Source Type Description
Recsite Point Willamette National
Forest Data Library
Point Locations of recreational sites
recorded at points
Recsite Poly Willamette National
Forest Data Library
Polygon Locations of recreational sites large
enough to be represented by area
features
Management
Areas
Willamette National
Forest Data Library
Polygon Categories of administrative areas as
defined by the 1990 Willamette
National Forest Plan
Private
Campgrounds
Google Earth,
Accompanying
websites
Point Locations of private campgrounds not
included in any of the previous data
layers, included in the analysis.

Two key data sets were used for the purpose of identifying areas that had frequent human
activity within the Willamette National Forest. One data set had all site types as points, and the
second had recreational sites as polygons. These data were available through the Willamette
National Forest Data Library. The USFS Geospatial Service and Technology Center created the
original data set. These data were created June 18, 2001 and were last updated June 28, 2010 so
this data is relatively concurrent with the fire data. The data were produced using one of two of
the following methods: 1) scanning a map or 2) manually digitizing a map by using a digitizing
table to capture the data. Point data depict developed recreational sites too small to be depicted at
1:24,000, polygon data depict sites large enough to be represented as polygons. The Data
Resource Management Division of the USFS Pacific Northwest Region later edited the data to
adjust certain locales within the Detroit Ranger District and divided the original data set into
separate point and polygon data sets for the purpose of spatial display and analysis.
30

Attribute data for both the point and polygon data used in this study consisted of
SITE_TYPE (type of site), NAME, DESCRIPTION, and OWNER. For the purpose of complete
and accurate coverage of all recreational sites within the Willamette National Forest, it was
decided that both point and polygon data would be used. It should be noted that more than one
recreational site type may be present within a single location, but each unique site type is
considered to be a single feature.
More recreational site data were found in the Management Area data set from the
Willamette National Forest Data Library. This data set was created April 13, 1994 and was last
updated January 22, 2010. This data set contains polygons of all classes of management areas
within the Willamette National Forest as described by the 1990 Forest Plan, later amended by the
1994 NW Forest Plan. For the purpose of complete data sets, all polygons within the
Management Areas data set with a category similar to other site type categories in the previously
discussed data sets were added to the other recreational site data. The Willamette National Forest
Supervisors Office created this data set for the purpose of managing natural resources. No
information was available on how the data set was created, thus its spatial uncertainty is
unknown.
Three private campgrounds that were not part of the original data were discovered during
the editing process of the recreational site data. Information about the campgrounds was obtained
via word of mouth. The campgrounds’ locations were investigated using Google Earth in order
to attain their lat/long coordinates so that points could be added to the layer. Attribute data for
these campgrounds were added manually.
31

3.2.3.1. Creation of the recreational site and human infrastructures data layers
The creation of the recreation site and human infrastructures data layers used in the
analyses involved recategorizing the site types, separating the data into separate layers and
combining features from several sources. Figure 5 shows the workflows, discussed in detail
below, that were used in the preliminary editing process on the different data sets.

Figure 5. Flowchart showing the Recsite and human infrastructures data preparation

In the point recreational site data (called Recsite point in Table 4), there were 18 different
types of recreational sites and administrative sites described in the site_type column of the
attribute table. With 171-point recreational sites (not counting administrative sites, resorts, or
32

summer homes), it was necessary to sort and consolidate the categories. Initial editing grouped
the types into nine different categories: campgrounds, day use areas, lookouts, snow parks,
boating, trail shelters, resorts, Forest Service Administrative areas, and ski areas. Categories for
sites that had a blank entry for SITE_TYPE attribute field were determined by reading the
DESCRIPTION attribute. The private campgrounds were added to the point layer.
Due to low counts of site locations in some categories, it was necessary to further
consolidate categories having few locations with similar categories. Separate layers were then
created for each of the different categories. Town, Community, Administrative site, Hotel,
Lodge, Resort, and residential/private features were removed from the Recsite classification and
were converted into three separate data sets henceforth referred to as human infrastructures.
Table 4 shows the consolidation and final naming of the five Recsite and two human
infrastructures layers. The same reclassification process was then followed for the Recsite
polygon data and those data were likewise separated into the same seven layers.
33

Table 4. Consolidation of Categories and Naming of Data Layers for Recreational Sites and
Human Infrastructures
Original Category Revised Category
Final Layer
Name
# of
features
Dataset
Collection
Boating Site Boating Site Boating Site 14
Recreational
sites
Campground Campground
Campground 256 Group Campground Campground
Horse Camp Campground
Lookout/Cabin Lookout
Trail Shelter 30
Observation Site Lookout
Trail Shelter Trail Shelter
Blank (trail shelter) Trail Shelter
Blank (lookout) Lookout
Picnic Site Day Use Area
Day Use Area 22
Day Use Area Day Use Area
Blank (Picnic Area) Day Use Area
Other Rec. Concession Day Use Area
Skiing, Down Hill Skiing Area
Snowpark Snowpark
Snowpark 13
Trailhead Snowpark
Hotel, Lodge, Resort Lodge/Resort Private/
Residential
31
Human
infrastructures
Private Summer Homes Private Summer Homes
Forest Service Facility F.S. Facility Administration
Area
13
Other Agency site Administrative Area
Towns Town
Town 6
Community Town
Total 385

Once the point data was cleaned up and separated into seven-point layers, it was
necessary to merge those with the polygon data in the original Recsite polygons. This was
accomplished first by buffering all points to create polygons. The point buffer size was
determined by calculating the average of size measurements of randomly selected recreational
sites taken from imagery from Google Earth. The sites that were measured included three to four
of each of the different categories of recreational sites. The measurements are intended to
express the average footprint size of each type of recreational site. This is used to account for the
recognition that the spatial influence of these sites on the landscape is spread over an area, not
34

just at a one-dimensional point. Buffer sizing that was used for the recreational sites is listed in
Table 5. The buffered points were then merged with the appropriate layer from the Recsite
polygon data and any buffered point features that matched with a polygon of the same category
were removed.
Table 5. Type Buffer Size
Type Buffer Radius
Boating Site 100m
Campground 250m
Trail Shelter 75m
Day Use 125m
Snowpark 100m
Finally, the Management Areas data set additionally provided administrative areas,
community, and private summer homes polygons. As a result, any administrative sites in this
data source were added to the administrative site layer, the private summer homes were added to
the residential/private layer and towns and communities were added to the towns layer. These
additional data layers are described further below.
3.2.4. Road Data
The Willamette National Forest offers access to many areas through a vast network of
roads. Different categories of roads, from paved highway, county, and agency access roads to
seasonal gravel maintenance and logging roads, allow vehicles of the appropriate type to
navigate areas where permissible within the WNF. These roads give people access to many of
the different recreational sites found throughout the WNF.
The road data were obtained in a polyline vector format. The USFS originally designed
the road data to show specific types of roads, their designated use, seasons of use, types of
vehicles allowed, and whether they are intended for the purpose of fire protection or
35

administrative use. The data were collected using one or both of two different methods: 1) on-
screen digitizing from a georeferenced topographic map, or image background, or 2) GPS field
collected positions. The data set was acquired from the FSGeodata Clearinghouse. The data was
last updated February 27, 2015. The Metadata gave no indication of updates for new roads had
been added to the data set since it was originally created. Initial editing of the data required
clipping the road layer to match the forest boundaries.
Road layer data were buffered to an average width determined by using 20 random
samples of road widths throughout the WNF. Road widths were measured using Google Earth
Ruler at points where both sides of a road section were visible and displayed a definite road
edge. The samples were chosen manually by zooming in to visually determine if road sides were
visible. Five road sections were chosen from each of the four ranger districts. Road sections
where the road was visually wider than normal, such as intersections or turn-outs were excluded.
The 20 road widths were averaged resulting in a width of approximately 5m (exact average was
4.75m), this equaled a buffer of approximately 2.5m. Figure 6 shows the editing processes for
the road data.

Figure 6. Road data editing process
36

It was later decided that, for visualization purposes, the state highways needed to extend
beyond the forest boundary to show a general direction to the neighboring major cities. Data
depicting the state highways for the entire state, created by the Oregon Dept. of Transportation
(ODOT), were found at ArcGIS Online. The state highway data were clipped at random
distances that extended beyond the forest boundary to show relative direction to major cities. The
ODOT highway data were not used for analysis purposes, only for reference and visualization
since the Willamette National Forest road layer already contained all WNF roads and highways.
3.2.5. Trail Data
The trail data used for this study was created by the USFS to provide public information
about trails, their locations, and characteristics. Trail attribute data used in this study included
NAME (when available). Trail data were clipped to the forest boundary. Data for this data set
were found in the FSGeodata Clearinghouse. The data set was created February 27, 2015. The
Metadata gave no indication of updates that new trails had been added to the data set since it was
originally created Individual Forests are responsible for managing road and trail data sets. Trail
data was collected in the same manner as road data.
The trail layer was buffered to 0.75 meters which gives total trail width of 1.5 meters
wide. This width was estimated from personal experience from hiking Forest Service trails in the
area.
3.2.6. Private, Residential, and Lodging Areas
Throughout the Willamette National Forest there are sections of land that are privately
owned. Some of these areas are simple forest plots, owned by people or businesses for forestry
harvest purposes. There are other areas where people have created small neighborhoods (too
small to designate them as towns or unincorporated communities) that are used as summer
37

homes, retreats, or resorts. For the purpose of this study, we focused on the areas of higher
density human activity, such as the residences and those areas that are used for retreats or
recreational use. Private land that was strictly used for forestry purposes was considered the
same as USFS lands that had been designated for forestry purposes. Access to the areas was
available using USFS roads and trails. By including the private lands as part of the WNF’s
extent, it made it possible to examine complete wildfire coverage through the entire WNF.
Private, residential, and lodging data includes any privately-owned lands within the
WNF. The residential type includes any form of private ownership of a house, condominium, or
place in which a human resides, and the lodging type includes any commercial residence where
individuals rent or pay for temporary residency. There were private land areas within the
Management Area data set that were homes that are owned privately, but the house resides on
Forest Service lands. These types of residence are known as a lease type residency for the
purposes of this study, were considered areas of private residences.
As explained above in Section 3.2.3, Private, Residential, and Lodging data were
extracted from the Management Areas data set created by the USFS at the Willamette National
Forest Headquarters. Polygons that were defined as private land ownership, residency, lodge, or
resort were separated out and merged with the same data type that had been extracted from the
original Recsite data. Attribute data used for this study included FP_NAME, FP_MGT_ALC
(Forest Plan Management Allocation), and MGT_ALC (Management Allocation Consolidation
of 1990 Plan and 1994 Plan).
3.2.7. Administrative Areas
Administrative areas were included to ensure all types of human activities found within
the WNF were considered. Administrative areas for the purpose of this study are areas that are
38

controlled by a government agency and are off-limits or have limited public access. Forest
Service Ranger Stations, Forest Service and Oregon Dept. of Transportation maintenance yards,
and fish hatcheries are a few examples. Agencies found within the Willamette National Forest
who maintain such areas other than the Forest Service are the Oregon Dept. of Transportation,
the Oregon Dept. of Parks and Recreation, Oregon Dept. of Fish and Wildlife, and Oregon Dept.
of Forestry.
As explained in Section 3.2.3, these data were found in the Management Areas layer. As
with the Private, Residential, and Lodging data, only the FP_NAME, FP_MGT_ALC (Forest
Plan Management Allocation), MGT_ALC (Management Allocation Consolidation of 1990 Plan
and 1994 Plan) and LABEL attributes were used.
3.2.8. Town Sites
Data for towns and communities was filtered from the Management Areas data set
discussed in previous sections of this chapter. The town and community polygons were identified
and extracted. Then the polygons were trimmed to the outermost boundary of built-up residential
areas visible in Google Earth. Thus, it was possible to confirm the functional extent of the towns
and communities and to verify that the included polygons did follow the outskirts of the towns.
Issues with the town polygon data were limited to the original polygons’ size and shape, which
included large private lands next to the towns. These private forestry lands did not contribute to
the towns as locations of focused human activity and as such were removed from the polygons to
limit the influence of towns or communities to their residential and commercial footprints. Only
the FP_NAME, FP_MGT_ALC (Forest Plan Management Allocation), MGT_ALC
(Management Allocation Consolidation of 1990 Plan and 1994 Plan) and LABEL attributes were
used from the data set. All towns and communities are positioned well within the forest
39

boundaries, but the data were clipped to the forest boundary extent to eliminate any possible
problems.
Within this layer are four polygons representing the towns of Detroit, Idanha, Westfir,
and Oakridge. There are also 2 polygons representing the unincorporated communities of
Rainbow and McKenzie Bridge. Originally, town and community data were not to be included in
this analysis given the decision initially to disregard the influence on wildfires of population
expansion into the Wildland Urban Interface discussed earlier. It was later decided that towns
and communities do not influence wildfires strictly through population expansion. Tourism and
major highways bring a large number of people through the towns and communities to
recreational areas nearby. These towns and communities contain businesses that attract tourists,
such as stores, restaurants, and gas stations. As well, these denser population areas may be
expected to have a small impact on the wildfire probability as human activities within a
residential setting do include a number of possible wildfire ignition types, such as barbeques,
fireworks, and backyard burning.
3.2.9. Summary of Data Layers Used in the Analyses
As a result of this data preparation, a total of 385 sites were included for further analysis.
Edge effects caused by clipping to the WNF boundary are assumed to be minimal due to the size
of the WNF and the small number of recreational sites and administrative areas near the edges of
the WNF. Figure 7 shows the locations of all the sites used in this study. Also, Appendix B
includes individual maps of all the different site types.
40

Figure 7. Locations of all recreational sites and human infrastructures included in this study.
Points were used to show polygon sites too small to see on this map. Their colors match the
relevant polygon colors shown in the legend.
41

Other features that may influence wildfire occurrences exist within the Willamette
National Forest. Lakes and streams, for example, are usually accessible by roads or trails and are
often located near a campground, day use area, a lodge or resort, private residence, or one of the
other types of features explored in this study. Thus, water features were not included in this study
since it was assumed that any human caused wildfire that occurred near a water feature would
also be associated with the presence other features previously described. Similar decisions were
made about many other possible features that could be included in this study.
3.3. Preliminary Data Exploration
Once the data layers were prepared, preliminary data exploration could begin. To start
exploring the relationship between wildfire occurrences and the recreational sites (campgrounds,
day use areas, trail shelters, snowparks, and boating areas) and human infrastructures
(administrative areas, residential/private, and towns), it was necessary to visualize the data
together. Initial data exploration used the Kernel Density tool to explore the spatial relationships
between wildfire ignitions and all recreational sites, roads, trails, administrative areas, and
towns/communities. This section describes how this tool was used and discusses the results of
this preliminary data exploration in preparation for the main analysis described in Chapter 4.
3.3.1. Kernel Density
Kernel density calculates the number of features or events per unit area within a region
defined by a search radius (O'Sullivan and Unwin 2014). Cell values decrease with distance from
locations with high numbers of features, until the cell value is zero where no features are found
within the search radius.
Kernel density was used to assess visually if any type(s) of wildfire occurrence appeared
to be more related to a particular type of site. It was assumed, for example, that wildfires caused
42

by campfires would occur more frequently near campgrounds. The density assessment compared
wildfires caused by Smoking and Campfires individually and all fires collectively against all site
types.
To insure consistent results, the same input parameters were used for all kernel density
analyses. Table 6 shows the input parameters for each field of the kernel density tool.
Table 6. Kernel Density Input Parameters
Input field Input Parameter
Input point or
polyline feature
Individual wildfire cause layer (Smoking, Campfire) or
the complete fire dataset
Population field None
Output raster Name based on cause type
Output cell size 500 meters
Search radius 5 kilometers
Area units Square kilometers (default when units are metric)
Output values are Densities
Method Planar

A 500 meter output cell size was based on an assumed distance a person might casually
wander from a site, road, or trail as well as the +/- 200 meter imprecision of fire locations. The
Search Radius was set to five kilometers allowing for a focused density result around clusters of
occurrences. Area unit input parameter defaulted to square kilometers based on the input feature
class linear units on meters. The Out_Cell_Values was set to Densities in order to look for high
areas of wildfire occurrences.
3.3.2. Results of Kernel Density Preliminary Data Exploration
The results of the preliminary data exploration revealed that areas of high fire density did
not correlate strongly with what was thought to be the site feature responsible for the occurrence.
For example, Figure 8 shows three copies of the same density surface created from fires caused
by campfires along with three different sets of the site features. These fires caused by campfires
43

did not show any particular spatial relationship with the locations of campgrounds and day use
areas, as hoped (Figure 8A). While there are campgrounds and day use areas in the areas of
higher density, there are plenty of campgrounds and day use areas in the areas of 0 density.
When compared with the locations of trails (Figure 8B) and roads (Figure 8C), the density results
for fires caused by campfires showed some correlation, but the results were not consistent
enough to suggest a strong relationship worth exploring statistically.

Figure 8. Density of campfire caused wildfires in relation to (A) campgrounds and day use areas,
(B) trails and residential/private, and (C) roads and towns
44

While it was anticipated that smoking caused wildfires would reveal a significant
relationship to day use areas, towns, and communities, the density analysis results shown in
Figure 9 do not suggest any particular association between areas of high density and any
particular site type. Significantly, the areas where high density of smoking related fires occurred
are also areas where there are many different types sites.

Figure 9. Density of smoking caused wildfires and (A) campgrounds and day use areas, (B) trails
and residential/private, and (C) roads and towns
45

Finally, Figure 10 shows the density surface for the full set of 493 human caused wildfire
ignition locations for the study period. As above, there is little evidence of strong spatial
relationships here, though there does seem to be some tendency for the densest areas to follow
the highway corridors. There is also a higher density in close proximity to the larger towns to the
north and the south.

Figure 10. Density of human caused wildfires and (A) campgrounds and day use areas, (B) trails
and residential/private, and (C) roads and towns
46

3.3.3. Overall Observations from Density Analysis
The results of the kernel density analysis revealed that wildfire occurrence frequencies
are not consistent with any specific recreation site type. Site types found throughout the WNF
may show a relationship to wildfires in certain areas, but not in others. Areas of high population
densities occurred in higher total wildfire density, especially near the towns of Detroit and
Westfir/Oakridge. Though the kernel densities are higher in these areas, so are the number of
people. Also, it must be noted that the areas near towns and communities generally have many
types of recreational sites nearby. This crowding of recreational sites, residential/private, and
other human infrastructure sites into small areas makes it difficult to find relationships between
specific wildfire types and site types. As for sites in less confined areas, the seemingly random
nature of wildfire and site relationships is inconclusive. The results of this preliminary
exploration are not substantial enough to suggest any type of relationship. Therefore, it was
decided to explore these relationships collectively through the use of a maximum entropy
technique as described in the next chapter.

47

Chapter 4 Data Mining with Maxent
This chapter focuses on the use of Maxent to predict the probabilities of wildfires as related to
the various recreational sites and human infrastructures mentioned in the previous chapters. In
this study, fire occurrences are the “species” presence data and recreational sites and human
infrastructures provide the environmental layers.
Most of the individual wildfire cause occurrences used in this study are small samples.
Maxent has the ability to model predictions with species locations that contain small sample
quantities (Pearson et al. 2007). Yet, to ensure valid results, for this study only individual
wildfire causes with more than 50 occurrences (i.e. Campfire and Smoking) were included in the
Maxent modeling. As well, a Maxent model using all wildfire occurrences collectively was
developed.
4.1. Using Maxent to Understand Relationships Between Fires and Sites
For each wildfire cause used as an input, a Maxent model was constructed. Maxent
provides, for each model calculated, a raster image of the probability surface as well as various
model statistics, response curves, analysis of variable contributions, and tables depicting the
jackknife testing result of the effect and importance that each environmental variable has on the
prediction. More than the probability surface, for the purposes of this study, the response curves
and variable contribution outputs are particularly useful in understanding the relationships
between recreational sites, human infrastructures and fire occurrences.
Maxent requires specific input formats for both the presence locations and the
environmental variables. The presence data input format must be in a Comma Separated Values
(CSV) format. The CSV file must contain three columns, the species name (for this study the
wildfire cause was used), an X-coordinate, and a Y-coordinate. These data simply give the
48

presence locations within the study area. The environmental data may be either categorical or
continuous—a mix of both types in a single model is acceptable—and it must be in ASCII raster
format with all layers having the same cell size and extent. Since all of the environmental layers
in this study began as vector data, all of it required conversion to raster format.
As noted above, Maxent accepts two different types of environmental data, categorical or
continuous. This study used both to 1) analyze the locations of wildfire occurrences with respect
to the specific location of various recreational sites (below called the “categorical data analysis”),
and 2) examine how distance affected the relationship between wildfire ignitions and recreational
sites (the “continuous data analysis”). The next two sections describe how both types of data
were prepared for use in Maxent.
4.2. Categorical Data Preparation
Initially, all the data used as environment variables in this study were categorical. All of
it needed to be converted to raster format. This section focuses on the methods and preparation
for the categorical data.
4.2.1. Wildfire Data
The species location data for the purpose of this study were the wildfire cause data
described in Section 3.2.2. The Wildfire Cause attribute table for the vector point file did not
contain X or Y coordinate values. To solve this issue, the Add XY Coordinate tool was used to
add X and Y Coordinates to the attribute table. The Add XY Coordinate tool simply adds two
fields to the attribute table POINT_X and POINT_Y, which are the calculated values of the
positions of the wildfire occurrences based on the coordinate system of the dataset. The wildfire
cause attribute table was then exported from ArcMap to an Excel file where the wildfire cause,
X-coordinate, and Y-coordinate columns were separated out and saved as a CSV file.
49

4.2.2. Forest Boundary Data
To ensure that all environmental variable layers were the same, the Forest Boundary layer
was used to set the extent of all rasters used for the environmental variables. The Forest
Boundary layer was converted from a vector format to a raster format. The cell size for the raster
was set at 500 meters, the distance a person might casually wander from a site, road, or trail.
Values for the raster that indicated the inside of the forest boundary were 1, and any values found
outside the forest boundary were No Data. The forest boundary layer was not an environmental
variable, but was used only to set the extent of the study area.
4.2.3. Recreational Site Data
Preparation of the recreational site data for use as an environmental variable required that
each recreational site type layer be prepared individually. Though each layer was prepared
individually, all of the processes were repeated with each layer. Each vector polygon layer (as
described in Section 3.2.3) was converted to a raster format with coded values to set the cell
values for the raster (e.g. campgrounds = 1, day use area = 2). Any cell that contained any part of
an original polygon was coded with the relevant code value. Cells that did not contain a
recreational site value were coded as No Data. The raster extent was set within the environments
options to that of the Forest Boundary, and the cell size was set at 500 m to match all other raster
layers since all Maxent environment layers must have the same extent and cell size. The rasters
were then converted to an ASCII file format.
4.2.4. Polyline Data: Road and Trail
The Road and Trail polyline vector data sets were each converted into separate raster
layers using the polyline to raster tool. The raster value was set during the raster conversion
using 1 = road or trail, and No data = no road or trail. Values for non-polyline raster cells were
50

coded as No Data. The raster extent was set to match that of the forest boundary and the cell size
for the raster layers was set at 500 m to match the recreational site raster layers. The Road/Forest
Boundary and the Trail/Forest Boundary raster layers were then converted to an ASCII format
for Maxent modeling.
4.2.5. Polygon Data: Private/residential, Town, and Administrative Areas
The Private/residential layer, Town layer, and the Administrative Areas layer were each
converted to a raster format using the Polygon to Raster tool. The rasters was coded during the
rasterization process using the site_value attribute (Towns = 7, Administrative areas = 8, and
Residential/Private = 9). Raster cells with no values were coded as No Data. Raster cell size for
each layer was set at 500 m to match the Forest Boundary layer. Raster extent was set within the
environment options to the forest boundary. Each raster layer was then converted to an ASCII
file format to be used in Maxent.
4.3. Continuous Data Preparation
To determine if distance was a factor in the relationships between wildfires, recreational
sites and human infrastructures, continuous data were used. The continuous data model used
raster layers that contain cell values representing the distances from the cell centers of the nearest
feature of interest, such as recreational sites or human infrastructure locations.
The rasters created for the categorical model were used as input into the Euclidean
Distance tool to create continuous data rasters for distance analysis in Maxent. Euclidean
Distance calculates the distance from cell center to neighboring cell centers by using the
Euclidean Distance algorithm, calculated as the length of the hypotenuse of a right triangle. The
categorical rasters were used as input because, according to the Euclidean Distance tool
description in ArcGIS Help, polygon layers with small polygons should be converted to rasters
51

prior to using Euclidean Distance because the internal raster process of Euclidean Distance uses
a cell-centered method. This method may inadvertently remove any polygon that is not located in
a cell center. The output cell size was set to 500 m. The extent was set in the environmental
options to the forest boundary raster layer described in Section 4.2.2. The rasters were then
converted to ASCII format using the ASCII to Raster tool.
4.4. Maxent Data Methods and Procedures
Maxent allows for a variety of input options, depending on the output desired. For this
study, it was only required that Maxent create models to show any probabilities of relationships
between wildfires and different types of recreational sites and human infrastructures. The
parameters chosen for the Maxent model runs are listed in Table 7. To allow easier interpretation
of results, the hinge and threshold features were not chosen. The threshold determines the
response level at which predicted presences would be indicated. It is not needed for this analysis.
Removing the hinge feature, which depends on the threshold setting, smooths out the response
curves while increasing uncertainty as variable values (i.e. distance) increase. In this analysis,
this is not important as low distance values are the values of interest.
Table 7. Parameter settings used for Maxent model runs
Maxent Parameter Input
Create Responsive Curve Yes
Do jackknife to measure variable importance Yes
Output format Logistic
Random Seed Yes
Replicates 50
Add all samples to background Yes

Response curves show how each environmental variable relates to the wildfire cause.
Jackknife testing was used to test each variable’s importance (Elith et al. 2011). The output
52

format chosen was Logistic, which is used to predict the probability of presence at all sites with
typical conditions for the species (or fire in this case) (Elith et al. 2011; Merow, Smith, and
Silander 2013). Random Seed was chosen so that a different test partition would be made with
each run ensuring that a different random subset of the background was used to test the data. The
modeling process was replicated 50 times since each run uses a different random seed and
multiple runs are necessary to converge on a stable model.
4.4.1. Presence and Environmental Variables Input Data
The environmental variables used for the Maxent modeling were campgrounds, day use
areas, trail shelters, towns, administrative areas, residential/private, roads, and trails. All
environmental variables were used in Maxent and the data type chosen (categorical or
continuous) depended upon whether an occurrence or a distance model was being run. The
wildfire CSV file (presence data) was also input into Maxent and the wildfire type(s) chosen to
be run in each model were selected.
The creation of five different models was attempted. Models runs were made for each of
campfire and smoking cause types using both the occurrence and distance environmental rasters.
The fifth model used all human caused wildfires reported within the Willamette National Forest
between 1995 to 2008 and was run using only the distance environmental rasters. Results of the
Maxent modeling effort are discussed in the next chapter.

53

Chapter 5 Maxent Results
Five models were created to explore the difference in the predictive capability of categorical and
continuous environmental data. Categorical data were used to model the effect of colocation of
recreational sites and human infrastructures with different types of wildfire ignitions. Continuous
data were used to explore if distances to features might give more meaningful results. The results
of each model are discussed below.
It is important to determine which variables are the most important to model. During each
step of the model training, Maxent keeps track of which variable is contributing the most to
fitting the model (Phillips 2006). For each iteration of the Maxent training, when a gain is
registered, it assigns the gain to the environmental variable that was responsible. This produces a
useful measure for assessing which variables are most important in determining presence.
5.1. Categorical Results
In hindsight, as might be expected, producing valid models using the categorical data was
not possible. Given that the recreational sites and human infrastructures are sparsely distributed
across the WNF, Maxent failed to find enough fire locations that aligned with any type of site
feature to produce the intended occurrence models. Only the distance models produced valid
results.
5.2. Continuous Data Results
This section explains the results from the distance modeling which can be used to
examine how distance from recreational sites and human infrastructures affect the presence of
wildfire ignitions. Maxent model results discussed here were gathered from the summary HTML
file produced during the modeling runs. These summary reports display the responses curves and
54

the environmental variable contributions resulting from the 50-fold cross-validation models run
for Willamette National Forest. Using the logistic output format option, Maxent creates two sets
of response curves to show how each environmental variable affects the Maxent output. The first
set of curves show that while all other environmental variables are kept at their average value,
the predicted probability of presence changes when each environmental variable is varied. The
second set of curves uses only the corresponding variable and reflect the dependence of predicted
suitability on both the selected variable and dependencies induced by correlations between the
selected variable and other variables. According to Phillips (2009), results for environmental
variables that are correlated may be misleading as the change in one variable will affect the
outcome of other variables. It is for this reason that Maxent creates the second set of response
curves as they are easier to interpret when there is correlation between variables.
A Maxent model report also displays an Analysis of Variable Contribution table. This
table lists the environmental variable, the percent contribution, and the permutation importance.
The percent contribution is the percentage of the gain that each variable contributes to fitting the
model (Phillips 2017). Again, when variables are correlated, the percent contribution should be
interpreted with caution as altering one variable might change the gain of another. The
permutation importance uses the AUC to determine the importance of the variable by measuring
increases or decreases in AU. Large decreases indicate that the variable is highly important to the
model. All values are normalized to be read as percentages.
As described in the previous chapter, the models discussed here are those for the
campfire and smoking caused fires as well as for the entire collection of human caused fires.
55

5.2.1. Campfire Caused Wildfire Model
There were 259 campfire caused wildfires throughout the Willamette National Forest
between 1995 to 2008. Maxent was able to produce a good model showing the predicted
probability of wildfire occurrences from the continuous data with an AUC of 0.773.
Most important for this analysis is consideration of how the different variables
contributed to the final model. Table 8 shows the percent contribution and permutation
importance of the variables in the model. The variables contributing the most to the model are
campground at 26.4 percent, residential/private at 17.4 percent, trail at 16.1 percent, and
snowpark at 9.2 percent. However, the variables in this study are highly correlated since, as
mentioned in Chapter 3, recreational areas often include more than one site type and these
contributions are impacted by the correlations between the site types. When considering
permutation importance, which shows how important the variable is when all other variables are
held constant, trail and snowpark ranked the highest at 15.9 percent, followed by
residential/private at 15 percent, and campground at 10 percent. All other variables had little
importance in the model.
Table 8. Variable contributions for campfire caused wildfires
Variable Percent contribution Permutation importance
campground 26.4 10
respvt 17.4 15
trail 16.1 15.9
snowpark 9.2 15.9
dayuse 8.2 3.4
boat 5.9 6.5
road 5.8 7.2
admin 5.1 12.9
trailshelter 3.1 6.4
town 2.8 6.9

56

Figure 11 shows the response curves from the jackknife test. Though the response curves
show results as far out as 60 kilometers, it is most important to focus on the results within three
kilometers from the source. This is assumed to be the maximum distance to which the risk of
ignition from human activities will extend from a given site or feature. The red vertical line in
the graphs represents approximately three kilometers. Though these graphs extend out tens of
kilometers, the extreme end of these graphs are not relevant as they are far beyond any distance
relevant to this study.
As expected, campgrounds (Figure 11A) showed a clear relationship with campfire
ignitions where probability of a wildfire ignition is high close to campgrounds and decreases
rapidly as the distance from campgrounds increases. Similar results were found with the trail
variable in that it showed the probability of an occurrence of a campfire caused fire is high close
to trails and decreases quickly with distance from trails. The day use, residential/private, and
snowpark area variables also showed a decrease in probability of occurrence with distance,
Figure 12 shows each of the variables in relationship to the model’s probability surface.
While the individual relationships are not strong, areas of high probability have combinations of
site types, such as campgrounds, day use areas, or campgrounds and boating areas.
57

Figure 11. Maxent model response curves for campfire caused wildfires in relation to (A)
campgrounds, (B) day use areas, (C) residential/private, (D) roads, (E) snowpark, and (F) trails.
Values on the vertical axes are probability of occurrence and on the horizontal access are meters.
Red vertical line indicates approximately three kilometers. Blue horizontal line at 0.5 probability.
58

Figure 12. Predicted campfire ignitions in relation to (A) campgrounds, (B) residential/private,
(C) day use areas, (D) trail, (E) administrative areas, and (F) boating areas.
59

5.2.2. Smoking Caused Wildfires
Within the Willamette National Forest there were 51 recorded smoking related wildfires
during the time period of this study. A Maxent model was created to look for predicted
probabilities of smoking caused wildfire occurrences. The model revealed some strong evidence
that certain environmental variables are correlated to smoking related wildfires with an AUC of
0.785.
Table 9 shows the percent contribution and the permutation importance of each variable.
The variable contributions for smoking related wildfires as modeled by Maxent showed that
residential/private and roads were the two main variables contributing to the model. The
residential/private variable had a 16.5 percent contribution and a 5.8 percent permutation
importance. Road had a 15.7 percent contribution and a 41.5 percent permutation importance.
Other variables also contributed substantially to the model with snowpark at 15.3 percent,
campground at 15 percent contribution, with an 18.6 percent permutation importance, day use at
13 percent, and town at 11 percent.
Table 9. Variable contributions for smoking caused wildfires
Variable Percent contribution Permutation importance
respvt 16.5 5.8
road 15.7 41.5
snowpark 15.3 6.7
campground 15 18.6
dayuse 13 3.5
town 11 10.7
admin 5.3 1.1
trail 4.4 10.9
boat 2.1 0.1
trailshelter 1.7 1.1

60

The Maxent model jackknife test response curves showed a reduction of occurrence
probability over distance in all variables (Figure 13). All the variables show a decrease in
probability over distance. Roads (Figure 13D) proved to be the variable with the most significant
decrease over the shortest distance suggesting that smoking related occurrences happen very
close to roads. The red vertical line in the graphs indicates approximately three kilometers, a
distance a person might easily walk from a recreational site, road, or trail.
Looking at Figure 14, it can be seen how campgrounds, day use areas, residential/private,
and towns in the northern half of the WNF (Figure 14A, B, C, and D) are located near the higher
areas of probability. Areas of high probability show a distinct correlation with recreational sites
and human infrastructures that are near the highway corridors. Comparing Figures 14A, B, C,
and D to the highway corridors (Figure 14F), it can be seen that the areas of higher probability
follow the highway corridors, except in a few areas. In Figure 14F, it is apparent that the road
network extends occurrence probability.

61

Figure 13. Maxent model response curves for smoking caused wildfires in relation to (A)
campgrounds, (B) day use areas, (C) residential/private, (D) roads, (E) snowpark, and (F) towns.
Values on the vertical axes are probability of occurrence and on the horizontal access are meters.
Red vertical line indicates approximately three kilometers. Blue horizontal line at 0.5 probability.

62

Figure 14. Predicted smoking ignitions with (A) campgrounds, (B) residential/private, (C) day
use areas, (D) towns, (E) snowpark, and (F) roads and highways.
63

5.2.3. Total Wildfires for the Willamette National Forest
The USFS recorded 493 human caused wildfires within the Willamette National Forest
between 1995 to 2008. This total amount includes all of the wildfires investigated within this
study as well as railroad and miscellaneous cause types which were removed prior to any other
investigation or modeling. The only change made to the data was a new csv file created from the
complete wildfire data layer. This csv file utilized the Forest name as the presence data instead of
wildfire categories. All environmental variable layers used in the campfire and smoking models
were used in this model and the same 50-fold cross-validation parameter set was used.
The variable contributions for total wildfires in the Willamette National Forest (Table 10)
saw campground with 27.6 percent, day use areas at 14.6 percent, and administrative areas at 11
percent. Others were under 10 percent. Permutation importance put snowparks at 23.8 percent,
roads at 19.2 percent, day use areas at 13 percent, and residential/private at 12.5 percent.
Table 10. Variable contributions for total wildfire in the Willamette National Forest
Variable Percent contribution Permutation importance
campground 27.6 13
dayuse 14.6 4.5
admin 11 3.2
respvt 9.6 12.5
snowpark 9.6 23.8
trail 9.1 7.1
road 7.1 19.2
boat 6.5 8.7
town 2.8 6.2
trailshelter 2.2 1.7

64

Figure 15. Maxent model response curves for smoking caused wildfires in relation to (A)
campgrounds, (B) day use areas, (C) administrative areas (D) residential/private, (E) trail, and
(F) roads. Values on the vertical axes are probability of occurrence and on the horizontal access
are meters. Red vertical line indicates approximately three kilometers. Blue horizontal line at 0.5
probability.
65

The total wildfire model Jackknife test response curves (Figure 15) were very similar to
those of the campfire model (Figure 11) when comparing them out to a distance of three
kilometers. This was not surprising given that campfires accounted for 52 percent of the total
wildfires reported. All variables showed a constant and rapid decrease out to the 3-kilometer
mark indicating that fire occurrence probability diminishes quickly away from all sites.
When looking at the maps in Figure 16, you can see that the high probability areas are
located near areas of multiple sites and infrastructures. In Figures 16A, B, C, and D, you can see
that when campgrounds, day use areas, administrative areas, residential/private, and snowparks
are located near each other, probabilities are higher than in areas where sites and infrastructures
are less clustered. It is interesting to see that, like smoking, the areas of higher collective
probability follow the highway corridors (Figure 16E) while other roads, which cover a
significant amount of territory, occur where lower probabilities were also modeled.
66

Figure 16. Predicted wildfire ignitions in relation to (A) campgrounds, (B) day use areas, (C)
administrative areas, (D) residential/private and snowparks, (E) roads and highways,
and (F) trails.

67

Chapter 6 Discussion and Conclusion
The goal of this study was to examine the relationship between the location of human
activities and the location of wildfire ignitions in order to gain some insight about the potential
locations of future wildfires. This thesis examined the relationship between different causes of
wildfires and various kinds of recreational sites and human infrastructures found within the
Willamette National Forest. It explored the hypotheses that (1) there would be more campfire
caused wildfires near campgrounds, (2) smoking caused wildfires would be more abundant near
day use areas, towns, and residential areas, and (3) wildfires would be more prominent in areas
with many recreational sites or near human infrastructures. The results of the models suggest a
number of ways in which the wildfire danger might be addressed.
It was assumed that campgrounds would show a higher spatial relationship to campfire
caused wildfires simply based on the amount of campgrounds found within the WNF. Knowing
that each campsite within a campground contained some sort of firepit, fireplace or barbeque
station, the possibilities of wildfires resulting from a campground was expected to show high
probabilities of occurrence. Though results from the Maxent model did suggest that
campgrounds were a significant factor in campfire caused wildfires, it was surprising to find that
other site types (residential/private, trail, snowpark, and day use) showed a higher than normal
relationship to campfire caused wildfires.
Wildfire caused by smoking is a serious issue, especially during the hot, dry days of
summer. It was originally assumed that smoking caused wildfires would be more abundant
around day use areas, towns, and residential areas. These three areas did show a high probability
of occurrence to wildfires caused by smoking, though it was unexpected to find that roads and
68

campgrounds also showed a high probability. These are areas that have a higher volume of
human activities, and as such these areas are more prone to wildfires caused by smoking.
Looking at the total collection of human caused wildfires, it was assumed that there
would be a spatial relationship to areas of higher human density or higher human activities. As
expected, the results of this study have shown that areas with large numbers of recreational sites
or human infrastructure contribute to an increased potential for wildfire ignitions. Trails and
roads that may have higher than normal traffic patterns (popular trails and roads to popular
destinations) increase the potential for wildfires as well.
6.1. Future Work
The study of spatial distribution of wildfires and their relationships with human activities
is a broad subject and can be studied in many ways. This thesis attempted to get an overall view
of whether the spatial correlation between humans and wildfires could be identified. A study of
trends in the numbers and characteristics of visitors involved in various recreational activities in
the WNF might provide additional information to further explore the relationship between
wildfires and human activities. An even deeper understanding might be teased out with detailed
weather information, though it is likely that data at the fine resolution that would be needed for
this could be found. Combining the human causes of wildfire ignitions with environmental
factors might prove to be beneficial in predicting overall fire risks. This combination might
include such variables as fuel load, aspect, slope, and even humidity.
6.2. Fires and Park Users
Wildfires are a severe problem throughout the United States and their impact is more than
just the acres and firefighting costs that the news reports. There is loss of wildlife habitat. Fire-
damaged lands may take years, maybe decades, before the land heals enough to grow anything of
69

economic value again. Within natural areas such as the WNF, tourism, hunting, fishing, and
hiking are a few of the many activities people undertake, and closures due to wildfires create
significant loss of income for the agencies and businesses that depend on revenue from such
activities. Understanding the dynamic relationship between park users and wildfires is an
important topic that merits continued exploration. It is hoped that this research has provided a
small contribution in that direction.

70

References
Almeida, Miguel, José Raul Azinheira, Jorge Barata, Kouamana Bousson, Rita Ervilha, Marta
Martins, Alexandra Moutinho et al. 2017. “Analysis of fire hazard in campsite areas.”
Fire technology 53, no. 2: 553-575.
Burke, Constance J. 1979. “Historic fires in the central western Cascades, Oregon.” Master’s
thesis, Oregon State University.
Calef, M. P., A. D. McGuire, and F. S. Chapin III. 2008. “Human influences on wildfire in
Alaska from 1988 through 2005: an analysis of the spatial patterns of human impacts.”
Earth Interactions 12, no. 1: 1-17.
Camp, Philip E., and Meg A. Krawchuk. 2017. “Spatially varying constraints of human caused
fire occurrence in British Columbia, Canada.” International Journal of Wildland Fire 26,
no. 3: 219-229.
Chas-Amil, M. L., J. P. Prestemon, C. J. McClean, and J. Touza. 2015. “Human-ignited wildfire
patterns and responses to policy shifts.” Applied Geography 56: 164-176.
Cole, David N. 1981. “Vegetational changes associated with recreational use and fire
suppression in the Eagle Cap Wilderness, Oregon: some management implications.”
Biological Conservation 20, no. 4: 247-270.
Cole, David N. 1996. Wilderness recreation use trends, 1965 through 1994. Research Paper.
INT-RP-488. U.S. Department of Agriculture, Forest Service, Intermountain Research
Station. https://www.fs.fed.us/rm/pubs_int/int_rp488.pdf.
Dale, Virginia H., Linda A. Joyce, Steve McNulty, and Ronald P. Neilson. 2000. “The interplay
between climate change, forests, and disturbances.” Science of the Total Environment
262, no. 3: 201-204.
Davis, Charles. 2001. “The West in flames: the intergovernmental politics of wildfire
suppression and prevention.” Publius: The Journal of Federalism 31, no. 3: 97-110.
Donovan, Geoffrey H., Peter Noordijk, and Volker Radeloff. 2008. “Estimating the impact of
proximity of houses on wildfire suppression costs in Oregon and Washington.” In
Proceedings of the Second International Symposium on Fire Economics, Planning, and
Policy: A Global View. General Technical Report PSW-GTR-208. U.S. Department of
Agriculture, Forest Service, Pacific Southwest Research Station.
https://www.fs.fed.us/psw/publications/documents/psw_gtr208en/psw_gtr208eng.pdf
Elith, Jane, Steven J. Phillips, Trevor Hastie, Miroslav Dudík, Yung En Chee, and Colin J. Yates.
2011. “A statistical explanation of Maxent for ecologists.” Diversity and Distributions
17, no. 1: 43-57.
71

Flannigan, Michael D., Brian J. Stocks, and B. Mike Wotton. 2000. “Climate change and forest
fires.” Science of the Total Environment 262, no. 3: 221-229.
Fox, D. M., Nicolas Martin, Pierre Carrega, Julien Andrieu, Cyriel Adnès, Karine Emsellem, O.
Ganga, F. Moebius, N. Tortorollo, and E. A. Fox. 2015. “Increases in fire risk due to
warmer summer temperatures and wildland urban interface changes do not necessarily
lead to more fires.” Applied Geography 56: 1-12.
Gebert, Krista. 2007. “Wild Fire Suppression Costs.” Montana Business Quarterly 45, no. 3: 14-
18.
Hawbaker, Todd J., Volker C. Radeloff, Susan I. Stewart, Roger B. Hammer, Nicholas S.
Keuler, and Murray K. Clayton. 2013. “Human and biophysical influences on fire
occurrence in the United States.” Ecological Applications 23, no. 3: 565-582.
Hirsch, Kelvin, Victor Kafka, Cordy Tymstra, Rob McAlpine, Brad Hawkes, Herman Stegehuis,
Sherra Quintilio, Sylvie Gauthier, and Karl Peck. 2001. “Fire-smart forest management: a
pragmatic approach to sustainable forest management in fire-dominated ecosystems.” The
Forestry Chronicle 77, no. 2: 357-363.
Lenart, Melanie. 2006. “Collaborative stewardship to prevent wildfires.” Environment: Science
and Policy for Sustainable Development 48, no. 7: 8-21.
Manning, Robert. 2001. “Programs that work. Visitor experience and resource protection: a
framework for managing the carrying capacity of National Parks.” Journal of Park and
Recreation Administration 19, no. 1: 93-108.
Massada, Avi Bar, Alexandra D. Syphard, Susan I. Stewart, and Volker C. Radeloff. 2013.
“Wildfire ignition-distribution modelling: a comparative study in the Huron–Manistee
National Forest, Michigan, USA.” International Journal of Wildland Fire 22, no. 2: 174-
183.
Merow, Cory, Matthew J. Smith, and John A. Silander. 2013. “A practical guide to Maxent for
modeling species’ distributions: what it does, and why inputs and settings matter.”
Ecography 36, no. 10: 1058-1069.
McKenzie, Donald, Ze'ev Gedalof, David L. Peterson, and Philip Mote. 2004. “Climatic change,
wildfire, and conservation.” Conservation Biology 18, no. 4: 890-902.
McLemore, Shannon. 2017. “Spatio-Temporal Analysis of Wildfire Incidence in the State of
Florida.” PhD diss., University of Southern California.
Miller, Jay D., H. D. Safford, Michael Crimmins, and A. E. Thode. 2009. “Quantitative evidence
for increasing forest fire severity in the Sierra Nevada and southern Cascade Mountains,
California and Nevada, USA.” Ecosystems 12, no. 1: 16-32.
72

Monahan, W. B., A. Rosemartin, K. L. Gerst, N. A. Fisichelli, T. Ault, M. D. Schwartz, J. E.
Gross, and J. F. Weltzin. 2016. “Climate change is advancing spring onset across the U.S.
national park system.” Ecosphere 7, no. 10.
National Interagency Fire Center. 2016. “Federal Firefighting Costs (Suppression Only).” U.S.
National Interagency Fire Center, Boise, ID.
https://www.nifc.gov/fireInfo/fireInfo_documents/SuppCosts.pdf.
ORWW. 2004. “B&B Complex Forest and Fire History.” Oregon Websites & Watersheds
Project, Inc. Accessed June 14, 2018.
http://www.orww.org/B&B_Complex/History_&_Research/index.html.
O'Sullivan, David, and David Unwin. 2014. Geographic Information Analysis. New Jersey: John
Wiley & Sons.
Oyarzun, Megan Christine. 2016 “Predicting Archaeological Site Locations in Northeastern
California’s High Desert using the Maxent Model.” Master’s thesis, University of
Southern California.
Parisien, Marc-André, Susan Snetsinger, Jonathan A. Greenberg, Cara R. Nelson, Tania
Schoennagel, Solomon Z. Dobrowski, and Max A. Moritz. 2012. “Spatial variability in
wildfire probability across the western United States.” International Journal of Wildland
Fire 21, no. 4: 313-327.
Pearson, Richard G., Christopher J. Raxworthy, Miguel Nakamura, and A. Townsend Peterson.
2007. “Predicting species distributions from small numbers of occurrence records: a test
case using cryptic geckos in Madagascar.” Journal of biogeography 34, no. 1: 102-117.
Phillips, S. J. 2017. A Brief Tutorial on Maxent. Available from url:
http://biodiversityinformatics.amnh.org/open_source/maxent/. Accessed on 2018-06-17.
Phillips, Steven J., Robert P. Anderson, and Robert E. Schapire. 2006. “Maximum entropy
modeling of species geographic distributions.” Ecological modelling 190, no. 3-4: 231-
259.
Pew, K. L., and C. P. S. Larsen. 2001. “GIS analysis of spatial and temporal patterns of human
caused wildfires in the temperate rain forest of Vancouver Island, Canada.” Forest
Ecology and Management 140, no. 1: 1-18.
Romero-Calcerrada, Raul, C. J. Novillo, J. D. A. Millington, and I. Gomez-Jimenez. 2008. “GIS
analysis of spatial patterns of human caused wildfire ignition risk in the SW of Madrid
(Central Spain).” Landscape Ecology 23, no. 3: 341-354.
Stein, Susan M., James Menakis, M. A. Carr, S. J. Comas, S. I. Stewart, Helene Cleveland,
Lincon Bramwell, and V. C. Radeloff. 2013. “Wildfire, wildlands, and people:
understanding and preparing for wildfire in the wildland-urban interface-a Forests on the
Edge report.” Gen. Tech. Rep. RMRS-GTR-299. Fort Collins, CO. US Department of
73

Agriculture, Forest Service, Rocky Mountain Research Station. 36 p. 299.
http://www.fs.fed.us/rm/pubs/rmrs_gtr299.pdf
Stephens, Scott L., Jason J. Moghaddas, Carl Edminster, Carl E. Fiedler, Sally Haase, Michael
Harrington, Jon E. Keeley et al. 2009. “Fire treatment effects on vegetation structure,
fuels, and potential fire severity in western US forests.” Ecological Applications 19, no.
2: 305-320.
Syphard, Alexandra D., Volker C. Radeloff, Jon E. Keeley, Todd J. Hawbaker, Murray K.
Clayton, Susan I. Stewart, and Roger B. Hammer. 2007. “Human influence on California
fire regimes.” Ecological applications 17, no. 5: 1388-1402.
Urness, Zach. 2018. “Teen Who Caused Eagle Creek Fire in Columbia River Gorge Sentenced,
Apologizes in Letter.” Statesman Journal. February 16, 2018.
https://www.statesmanjournal.com/story/news/crime/2018/02/16/teenager-who-caused-
columbia-river-gorge-eagle-creek-fire-sentenced-apologizes-letter/346003002/.
U.S. Census Bureau. 2010. “Interactive Population Map.” May 05, 2011.
https://www.census.gov/2010census/popmap/.
U.S. Forest Service. 1990. Land and resource management plan: Willamette National Forest.
U.S. Dept. of Agriculture, Forest Service, Pacific Northwest Region.
http://hdl.handle.net/1794/6528.
U.S. Forest Service. 2014. “National Visitor Use Monitoring Survey Results: U.S. Forest Service
National Summary Report. Data collected Through FY 2014.” U.S. Department of
Agriculture. Forest Service. Washington D.C.
https://www.fs.fed.us/recreation/programs/nvum/
U.S. Forest Service. 2016a. Forest Service Fire Statistics System (FIRESTAT) User Guide. U.S.
Dept. of Agriculture, Forest Service. Washington D.C. https://fam.nwcg.gov/fam-
web/firestat/FIRESTATUserGuide.pdf.
U.S. Forest Service. 2016b. “Wildland Fire & Fuel.” U.S. Forest Service Research &
Development. February 18, 2016. https://www.fs.fed.us/research/wildland-fire/.
U.S. Forest Service. n.d. “About the Forest.” Willamette National Forest. Accessed June 22,
2018. https://www.fs.usda.gov/main/willamette/about-forest.
USGS. 2018. "The Public Land Survey System." The National Atlas of the United States of
America. Accessed June 28, 2018.
https://nationalmap.gov/small_scale/a_plss.html#links.
Westerling, Anthony L., Alexander Gershunov, Timothy J. Brown, Daniel R. Cayan, and
Michael D. Dettinger. 2003. “Climate and wildfire in the western United States.” Bulletin
of the American Meteorological Society 84, no. 5: 595-604.
74

Westerling, Anthony L., Hugo G. Hidalgo, Daniel R. Cayan, and Thomas W. Swetnam. 2006.
“Warming and earlier spring increase western US forest wildfire activity.” Science 313,
no. 5789: 940-943.
Yang, Jian, Hong S. He, Stephen R. Shifley, and Eric J. Gustafson. 2007. “Spatial patterns of
modern period human caused fire occurrence in the Missouri Ozark Highlands.” Forest
science 53, no. 1: 1-15.
Young, Alanna. 2014. “Analysis of Spatiotemporal Variations in Human-and Lightning-caused
Wildfires from the Western United States (1992-2011).” Master’s thesis, University of
Oregon. http://hdl.handle.net/1794/18702.

75

Appendix A. Locations of Individual Wildfire Causes

Figure 17. Arson related wildfires

Figure 18. Children related wildfires

Figure 19. Campfire related wildfires

Figure 20. Debris Burning related wildfires
76

Figure 21. Railroad related wildfires

Figure 22. Equipment use relate wildfires

Figure 23. Miscellaneous related wildfires

Figure 24. Smoking related wildfires

77

Appendix B: Recreational Sites and Human Infrastructures

Figure 25. Campground

Figure 26. Day Use Area

Figure 27. Trail shelter

Figure 28. Snowpark
78

Figure 29. Boating areas

Figure 30. Towns and communities

Figure 31. Adminstrative areas Figure 32. Roads
79

Figure 33. Residential/private

Figure 34. Trails

Abstract (if available)

Abstract Across the nation wildfires in national forests and parks annually affect millions of acres of public lands, destroying recreational sites, historical areas, and scenic wilderness, and costing taxpayers hundreds of millions of dollars every year in suppression costs and lost resources. This research examined the spatial correlation between human activities and human caused wildfire occurrences within the Willamette National Forest to explore whether these activities might be responsible for many wildfire ignitions. Between 1995 to 2008, 493 human caused fires occurred. The density of these fires was investigated to identify clustering near recreational sites and human infrastructures. Maxent was used to model the probability of wildfire occurrences in relation to the recreational sites and human infrastructure areas located throughout the Forest. ❧ It was initially hypothesized that more wildfires occur near specific kinds of recreational sites than elsewhere. Preliminary data exploration showed high densities of wildfire occurrences near the towns, human infrastructures, and major highways although these were also areas of clusters of recreational sites. Thus, it was not possible to identify visually which particular activities were most strongly related to wildfire ignitions. Maxent results revealed that areas of high population densities and recreation site clusters were more likely to correspond to areas of more human caused wildfire ignitions.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Modeling burn probability: a Maxent approach to estimating California's wildfire potential

PDF

Building better species distribution models with machine learning: assessing the role of covariate scale and tuning in Maxent models

PDF

Questioning the cause of calamity: using remotely sensed data to assess successive fire events

PDF

Using aerial imagery to assess tropical forest cover surrounding restoration sites in Costa Rica

PDF

The wildland-urban interface in Lassen County, California: a change analysis 2000–2015

PDF

A Maxent-based model for identifying local-scale tree species richness patch boundaries in the Lake Tahoe Basin of California and Nevada

PDF

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

PDF

Predicting post-wildfire regreen rates: an application of multi-factor regression modeling

PDF

A comparison of GLM, GAM, and GWR modeling of fish distribution and abundance in Lake Ontario

PDF

Species distribution modeling to predict the spread of Spartium junceum in the Angeles National Forest

PDF

Assessing the transferability of a species distribution model for predicting the distribution of invasive cogongrass in Alabama

PDF

Predicting archaeological site locations in northeastern California’s High Desert using the Maxent model

PDF

Evaluating predator prey dynamics and site utilization patterns of golden eagles using resource selection modeling and spatiotemporal pattern mining

PDF

Spatio-temporal analysis of Western snowy plover nesting at Vandenberg Air Force Base

PDF

Wetland mapping and restoration decision making using remote sensing and spatial analysis: a case study at the Kawainui Marsh

PDF

Spatial distribution of the Nile crocodile (Crocodylus niloticus) in the Mariarano River system, Northwestern Madagascar

PDF

Habitat suitability modeling of Mexican spotted owl (Strix occidentalis lucida) in Gila National Forest, New Mexico

PDF

Building a geodatabase design for American Pika presence and absence data

PDF

A spatial analysis of violent crime cold spots: testing the capable guardian component of the routine activity theory

PDF

Testing LANDIS-II to stochastically model spatially abstract vegetation trends in the contiguous United States

Asset Metadata

Creator Dickey, William P. (author)

Core Title Spatial analysis of human activities and wildfires in the Willamette National Forest

School College of Letters, Arts and Sciences

Degree Master of Science

Degree Program Geographic Information Science and Technology

Publication Date 09/26/2018

Defense Date 09/25/2018

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

Tag human activity,Maxent,OAI-PMH Harvest,wildfires,Willamette National Forest

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Kemp, Karen (committee chair), Longcore, Travis (committee member), Loyola, Laura Cyra (committee member)

Creator Email lostboy2127@gmail.com,wpdickey@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-74425

Unique identifier UC11671424

Identifier etd-DickeyWill-6783.pdf (filename),usctheses-c89-74425 (legacy record id)

Legacy Identifier etd-DickeyWill-6783.pdf

Dmrecord 74425

Document Type Thesis

Format application/pdf (imt)

Rights Dickey, William P.

Type texts

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

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

Repository Name University of Southern California Digital Library

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