WILDFIRE FUEL MAPPING USING AIRBORNE LASER SCANNING DATA:
CLIMATE ADAPTATION PLANNING WITH THE XÁXLI’P COMMUNITY
by
Patrick Robinson
B.Sc., University of Victoria, 2019

THESIS SUBMITTED IN PARTIAL FULFILLMENT OF
THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF SCIENCE
IN
NATURAL RESOURCES AND ENVIRONMENTAL STUDIES

UNIVERSITY OF NORTHERN BRITISH COLUMBIA
April 2025
© Patrick Robinson, 2025

Abstract
As climate change drives more frequent and severe wildfires, accurate forest fuel
mapping is critical for hazard assessment and adaptation planning. This study combines highresolution Airborne Laser Scanning (LiDAR) data with machine learning to produce detailed
fuel maps for the Xaxli'p Survival Territory in British Columbia. Using Random Forest
models, we mapped fuel layers prioritized by the Xaxli'p community, providing more precise
and locally relevant data than conventional wildfire fuel classifications. To train the models,
we compared two types of field data, structured, measurement-based samples and
community-based visual estimates, evaluating their effectiveness in predicting fuel
distribution. The findings show that integrating remote sensing with local knowledge can
improve wildfire hazard mapping in a local context. To support practical use, we developed
an interactive web map tailored for the Xaxli'p community to inform wildfire mitigation and
climate adaptation efforts. This research offers a scalable, community-driven approach to
wildfire hazard assessment and land stewardship.

ii

Table of Contents

Abstract

ii

Table of Contents

iii

List of Tables

vi

List of Figures

vii

Acknowledgements

xii

Prefacexiv
Chapter 1: Community-Directed Wildfire Risk Mapping and Management With the Xáxli'p 1
Research Foundations and Ethics
1
Research Partnership Background and Context
3
Project Objectives and Research Questions
8
The Xáxli'p Survival Territory and Landscape Restoration History
10
The Xáxli'p Community Forest Agreement
16
Xáxli'p Cultural Continuity
19
Conclusion
21
Chapter 2: Integrating Airborne Laser Scanning Data and Random Forest Modelling for
Wildfire Forest Fuel Mapping in the Xáxli’p Survival Territory
Introduction
Methods
Study Area
Airborne Laser Scanning (ALS) Data
Field Sampling Framework and Campaign
Techniques, Equipment, and Data Collection
Data Processing
Random Forest Modeling
Correlation Analysis of Model Outputs
Aggregate Fuels and Difference Maps
Results
Key Fuel Maps
Random Forest Model Performance
Correlation Between ALS Generated Fuels and FBP Fuel Types.
Aggregate Fuels and Difference Maps
Difference Maps
Discussion
Key Fuel Maps

24
24
27
28
31
32
36
38
40
44
46
48
48
51
51
61
64
68
68
iii

Random Forest Model and ALS Performance
Correlation Between ALS generated Fuels and FBP Fuel Types
Aggregate Fuels and Difference Maps
Aggregate Maps
Difference Maps
Conclusion

69
70
74
74
78
82

Chapter 3: Comparing Sampling Techniques in Forest Fuel Mapping: Structured Empirical
vs. Opportunistic Ocular Estimates with ALS and Random Forest
86
Introduction
86
Methods
89
Study Area, Data, and Fuel Layers
89
Structured Empirical (SE) Sampling
92
Ocular Estimates (OE) Sampling
93
Random Forest Modeling and Validation
98
Analysis
100
Results
102
Sample Coverage Assessment
102
Model Performance
105
Within-Model Comparison
105
Cross-Model Comparison
109
Fuel Maps Comparison
112
Discussion
114
Sample Coverage Assessment
114
Model Performance
117
Within-Model Comparison
120
Cross-Model Comparison
122
Fuel Maps Comparison
124
Conclusion
126
Chapter 4: Integrating Local Knowledge and Technology: The Development of a
Community-Based Mapping Tool for Xáxli'p Land Stewardship
The Evolution of a Community-Directed Research Partnership
Key Findings
Development of the Web Mapping Tool
An Evolving Story
Conclusions
Final Acknowledgements

129
129
135
138
141
145
148

References

150

Appendix
Appendix 1
Appendix 2
Appendix 3

169
169
172
178

iv

v

List of Tables
Table 1. Summary of Model Variables.................................................................................... 41
Table 2. Cross-validation R² values used for weighted aggregation....................................... 47
Table 3. Summary of performance metrics for each forest fuel type and additional forest
attributes...................................................................................................................... 51
Table 4. Cross-validated R² values for Random Forest models trained on SE and OE data for
predicting Canopy Fuel, Ladder Fuel, and Understory Fuel layers. ......................... 105
Table 5. Comparison of Structured Empirical (SE) and Opportunistic Ocular Estimate (OE)
Methods for Forest Fuel Mapping ............................................................................ 127

vi

List of Figures
Figure 1. The Xáxli'p Community Forest Crew. ....................................................................... 1
Figure 2. View from the valley bottom looking into a farm field in the Fountain Valley. ........ 2
Figure 3. Map illustrating the location of the study area within British Columbia, Canada,
with the main map highlighting the Xáxli’p Survival Territory in regional context and
an inset on the left providing a detailed outline of the Survival Territory boundary. ... 4
Figure 4. Aerial view of a forest plot surveyed during the 2020 field sampling campaign in
the northeastern section of the Xáxli'p Survival Territory. This image shows the
higher density fuel types found in the higher elevation areas of the Survival Territory
and was captured by drone............................................................................................ 5
Figure 5. Open field near Kwotlenemo Lake (Fountain Lake). Illustrating the
recommendation from Bezzola’s work regarding managing grasslands to avoid fire
spread to surrounding high-hazard forested areas. ....................................................... 6
Figure 6. The XCF crew and the UNBC research team discussing mapping and fieldwork
planning within the Xáxli'p Survival Territory. ............................................................ 9
Figure 7. View looking north over the south end of Kwotlenemo Lake in the central Fountain
Valley showing the high-density, connected fuels throughout the valley. .................. 12
Figure 8. Entrance to the Xáxli'p First Nation Fountain Valley community. The Xáxli'p First
Nation sign reads: “Be Fire Smart Aware”. ................................................................ 13
Figure 9. A wildfire risk reduction fuel-treated area near Kwotlenemo Lake showing the
treated area (right) next to the untreated area (left), highlighting the notable difference
in fuel loads. ................................................................................................................ 15
Figure 10. An ALS generated forest fuel load map displayed on an iPad in the field within a
fuel-treated area of the Xáxli'p Survival Territory. The map indicates areas of high
fuel density in red and low density in yellow and green, with the blue dot representing
the user’s location. The area on the right shows a dramatic increase in understory and
ladder fuel directly resulting from fire suppression. ................................................... 15
Figure 11. Harvested areas in the high-elevation northwestern portion of the Xáxli'p Survival
Territory. These were replanted using both conventional and experimental silviculture
techniques. .................................................................................................................. 16
Figure 12. Timeline of events leading up to the Xáxli’p Community Forest (XCF) Agreement
between the Xáxli’p community and the B.C. Ministry of Forests (MoF)” (Diver,
2016). .......................................................................................................................... 17
vii

Figure 13. Map showing the boundary of the Xaxli’p Survival Territory (orange) overlaid on
the boundary of the Xaxli’p Community Forest (blue). ............................................. 18
Figure 14. Xáxli'p Community Forest Crew Leader, exchanging a quiet moment with a horse
while journeying through an open field during the field data collection campaign. .. 20
Figure 15. Map illustrating the location of the study area within British Columbia, Canada.
The main map highlights the Xáxli’p Survival Territory, within the context of British
Columbia. The inset on the left provides an outline of the Survival Territory
boundary. ..................................................................................................................... 28
Figure 16. Map depicting the extent of the Xaxlip Survival Territory study area. Blue dots
represent the distribution of the 104 sample locations across the territory. Sample
locations were restricted to within 200 meters of roads, which influenced their
distribution throughout the territory. ........................................................................... 35
Figure 17. Diagram of forest plot layout showing a large 11.28m radius plot with four smaller
2.99m radius sub-plots positioned at 0, 90, 180, and 270 degrees. ............................ 37
Figure 18. A side-by-side display of the final output of the three key predicted fuel layers that
have been the focus of this analysis. ........................................................................... 49
Figure 19. A side-by-side display of the final output of additional predicted fuel layers that
have been included in the aggregate fuel layers calculation. ...................................... 50
Figure 20. Map of fuel types in the Xáxli’p Community Forest (Fountain Valley). Based on
the Canadian Forest Fire Danger Rating System (FBP System) Fire Behavior
Prediction (FBP) fuel types BC layer from 2019, provided by British Columbia
Wildfire Service (BCWS) Geomatics through a data-sharing agreement. ................. 53
Figure 21. Relative area of each fuel type, based on the Canadian Forest Fire Danger Rating
System (FBP SYSTEM) Fire Behaviour Prediction (FBP) fuel types BC layer from
2019. Provided by British Columbia Wildfire Service (BCWS) Geomatics through a
data-sharing agreement. .............................................................................................. 54
Figure 22. Correlation between canopy fuel and ladder fuel across 1000 random samples per
FBP Fuel Type............................................................................................................. 56
Figure 23. Correlation between canopy fuel and understory fuel across 1000 random samples
per FBP fuel type. ....................................................................................................... 58
Figure 24. Correlation between ladder fuel and understory fuel across 1000 random samples
per FBP Fuel Type. ..................................................................................................... 60
Figure 25. Normalized unweighted and weighted aggregated fuel layer 1. This map displays
the spatial distribution of the unweighted aggregated fuel layer 1, combining
normalized values for Canopy Bulk Density (CBD), Median Canopy Base Height
(CBH), Ladder Fuel, and Understory Fuel. ................................................................ 62
viii

Figure 26. Normalized unweighted and weighted aggregated fuel layer 2. This map displays
the spatial distribution of the unweighted aggregated fuel layer 2, combining
normalized values for Canopy Bulk Density (CBD), Median Canopy Base Height
(CBH), Ladder Fuel, Understory Fuel, and Stems per Hectare. ................................. 63
Figure 27. Difference map between canopy and ladder fuel layers. ....................................... 65
Figure 28. Difference map between canopy and understory fuel layers. ................................ 66
Figure 29. Difference map between ladder and understory fuel layers. ................................. 67
Figure 30. Map of sample locations from all three sampling campaigns. This includes
October and November OE sample campaigns and SE method samples, highlighting
spatial distribution and coverage achieved. ................................................................ 95
Figure 31. Map showing the distribution of SE and OE sample locations within the Xáxli’p
Community Forest/Survival Territory. Red dots represent the 695 OE samples, and
blue dots represent the 104 SE samples. ................................................................... 104
Figure 32. Within-model comparison of predicted versus observed fuel loads for Canopy
Fuel (top), Ladder Fuel (middle), and Understory Fuel (bottom)............................. 106
Figure 33. This figure displays cross-model comparisons for Canopy Fuel (top), Ladder Fuel
(middle), and Understory Fuel (bottom) layers. In each plot, the left side shows a
violin plot with individual values represented as jittered blue dots for the predicted SE
values against observed OE values, while the right side shows predicted OE values
against ALS-derived observed SE values. A red dashed 1:1 line serves as a reference
for perfect prediction accuracy, with a solid black line of best fit illustrating the trend
in the data. Pearson’s correlation coefficients and R² values indicate the predictive
performance for each fuel type across both methods. ...............................................110
Figure 34. Comparison of Canopy, Ladder, and Understory Fuel Layers and Their
Differences Between Structured Empirical and Ocular Estimates. This figure displays
fuel load predictions for Canopy Fuel (top), Ladder Fuel (middle), and Understory
Fuel (bottom), with maps derived from SE data (left), OE data (center), and a
difference map (right) for each fuel layer. The difference maps highlight areas where
SE and OE estimates diverge, with blue indicating areas where SE predictions are
higher than OE, turquoise representing moderate negative differences (OE lower than
SE), black showing no difference between methods (OE ≈ SE), orange indicating
moderate positive differences (OE higher than SE), and red highlighting areas where
OE predictions are higher than SE. This consistent color scheme emphasizes spatial
patterns and discrepancies between the two methods. ...............................................113
Figure 35. View of a switchback on Highway 99 and the east end of Seton Lake, just west of
Lillooet. ..................................................................................................................... 129

ix

Figure 36. Xáxli'p Community Forest Crew Leader, taking a measurement during field
sampling on a remote hillside. Xáxli'p Survival Territory, with Fountain Peak in the
background. ............................................................................................................... 131
Figure 37. Researcher Patrick Robinson measuring tree height with a hypsometer while
holding an iPad for data entry. This was during the fuel sampling campaign in the
Xáxli'p Survival Territory. The yellow measuring tape marks the plot diameter. .... 132
Figure 38. Aerial view of the transition between a fuel-treated area (left) and a non-fueltreated area (right) of the forest near Kwotlenemo Lake. ......................................... 133
Figure 39. View looking north up the Fraser River Valley from a large, old burned area in the
northeastern portion of the Xáxli'p Survival Territory. ............................................. 135
Figure 40. Xáxli'p Community Forest crew leaders gathered around the web mapping tool in
the Xáxli'p Community Forest office during a fieldwork planning session. ............ 136
Figure 41. Field data collection equipment used for the structured empirical (SE) data
collection. The equipment includes a drone for airborne imaging, navigation, and
situational awareness; a hypsometer for measuring tree heights; a differential GPS
unit for accurate plot center locations; a measuring tape for determining plot
diameters; an iPad with EpiCollect for streamlined data collection; and a diameter
tape for measuring tree DBH within plots. ............................................................... 137
Figure 42. Pouches of tobacco wrapped around a juniper tree, coincidentally discovered
during field sampling. This site, with a stunning view of Fountain Peak and the
expanse of the Xáxli'p Survival Territory, added an unexpected moment of reflection
to the day’s work. ...................................................................................................... 138
Figure 43. Screenshot of the web mapping interface showing a zoomed-in view of the
northeastern section of the Xáxli'p Survival Territory with the canopy fuel layer and
roads layer turned on. The web map allows users to explore all fuel layers and spatial
data in any web browser as long as they have an internet connection. In this view, the
canopy fuel layer highlights high fuel density areas in red, medium density in
orange/yellow, and lower density in green to dark green. The harvested areas (cut
blocks) in the northeastern portion of the study area stand out as green patches among
higher-density yellow, orange, and red areas, reflecting their lower canopy fuel load.
.................................................................................................................................. 139
Figure 44. A view of the Marble Canyon face from just north of the Xáxli'p Survival
Territory. The view was captured during the first field sampling campaign, with
yellowing larch trees in the foreground. ................................................................... 140
Figure 45. View of the entrance to the Xáxli'p Community, looking up at Fountain Peak. This
photo was taken during the early days of the COVID-19 pandemic, with the Xáxli'p
First Nation sign in the foreground reading "Closed to the public, residents only." 142

x

Figure 46. Two female mule deer encountered during the first field sampling campaign in the
Xáxli'p Survival Territory. ........................................................................................ 143
Figure 47. A view of the Xáxli'p Community at the entrance of the Fountain Valley, with
Fountain Peak towering overhead. This photo was taken from the north side of the
Fraser River Valley, looking south into the Fountain Valley and the northwestern
portion of the Xáxli'p Survival Territory. .................................................................. 144
Figure 48. Airborne view taken by drone over a sample plot during the initial field sampling
campaign. This photo, captured in the higher-elevation northeastern portion of the
Xáxli'p Survival Territory, shows a transition between three distinct forest types:
dense lodgepole pine on the left, a large, dense patch of spruce in the center, and a cut
block densely replanted with young pine and spruce. .............................................. 145
Figure 49. Researcher Patrick Robinson (me), loading truck with field gear after collecting a
sample plot during the field data collection campaign in the Xáxli'p Survival
Territory. Thanks to the Xáxli'p Community Forest team for providing the magnetic
Xáxli'p Community Forest decal seen on the side of Patrick’s truck used for access
during fieldwork around the territory. ....................................................................... 148

xi

Acknowledgements
To my family, thank you for always having my back, and for being my foundation.

To all my friends and colleagues, who’ve kept me sane, laughing, and smiling day by day
and year by year, your companionship has been the glue holding everything together.

To Annie Pumphrey, for your support, encouragement, friendship, love, and the lessons
learned along the way. Your presence was a beacon of light for me throughout this journey.

To Scott and Che, for your consistent support, encouragement, patience, and mentorship.

To Linda and Laurie, for helping to make my winter experiences in the north so
unforgettable. Your kindness and hospitality made those cold months warm and welcoming.

To the Pacific Institute of Climate Solutions, thank you for the funding that made this work
possible.

To the University of Northern BC, for bringing together a fantastic group of students, many
of whom became good friends during my winters in Prince George. Thank you also for the
financial support that enabled me to travel and present my research at the International
Conference of Forest Fire Research in Portugal.

xii

To the International Association of Wildland Fire, for supporting my journey to present my
research at the International Conference on Fire Behavior and Fuels in Ireland. Your support
opened doors I never thought possible.

To Dev Khurana, who, despite parting ways with the project under less-than-ideal
circumstances, was a true friend and played a important role in my learning experience.

And last, but certainly not least, to the Xáxli'p community, Ed, Derek, Rob, Karen, Puq,
Josiah, and the entire Xáxli'p Community Forest team. Thank you for your enthusiastic
engagement and guidance through this project. You have taught me so much, both as a
researcher and as a human. Your wisdom, warmth, and welcome have been transformative.

Thank you all.

xiii

Preface
This MSc thesis has been significantly shaped by generative AI tools, which have
played a key role in my research, development, and refinement since their public release in
November 2022.
I extensively used OpenAI’s ChatGPT (versions 3, 3.5, 4, 4o, 4.5, etc.) for
brainstorming, articulating concepts, generating and organizing content, refining text, and
assisting with coding. Other AI tools, including Google’s AI suite, Claude, Gemini,
Perplexity, Notebook LM, X AI’s Grok, and Deepseek, provided additional insights and
coding support. GitHub Copilot and Cursor enhanced my coding with real-time suggestions
and debugging, while Grammarly’s AI improved clarity and coherence. Scite assisted in
evaluating sources and verifying citations. These tools collectively supported my research,
analysis, and scientific methodologies.
Despite AI’s assistance, I maintained full control over all decisions, ensuring factual
accuracy, academic rigor, and data integrity. I fact-checked information, reviewed all code,
and upheld professional ethics throughout. While AI accelerated my learning and problemsolving, the intellectual and creative essence of this work remains my own.
AI’s rapid evolution raises ethical concerns, particularly regarding bias, data privacy,
and societal impacts. These issues are especially critical in Indigenous and community
contexts, where AI’s role must align with cultural values, sovereignty, and ethical
governance. Considerations around data ownership and the responsible application of AI in
research are paramount.
Since adopting these tools, I have navigated their ethical use with careful judgment,
even before formal institutional guidelines were available. As UNBC’s policies emerged, I
xiv

ensured compliance with academic integrity standards. This preface and acknowledgment of
AI use is intended to serve as a blanket statement that generative AI technologies, as outlined
above, have been extensively utilized throughout all aspects and stages of this work.
Exploring AI’s integration into my research has been a defining learning experience
of my graduate studies. I have gained skills that didn’t exist when I began this degree, skills
that will be invaluable in the years ahead. This project has provided a unique opportunity to
engage deeply with this new and transformative technology, enhancing my MSc journey in
profound ways.

xv

Chapter 1: Community-Directed Wildfire Risk Mapping and Management With the
Xáxli'p

Figure 1. The Xáxli'p Community Forest Crew.

Research Foundations and Ethics
Navigating contemporary research needs while preserving respect for Xáxli'p
community traditions and knowledge has been the most important aspect of this project.
From its inception, the project has been a Xáxli'p community-directed endeavor with the
central goal of generating information, tools, solutions, knowledge, and internal community
capacity directly useful to the Xáxli'p community. The partnership has been committed to
respecting ethical practices that prioritize the community's needs and cultural values. This
has been achieved by ensuring continuous engagement with the community throughout the
research process, from setting research directions, performing data collection, and
developing, validating, and applying the maps toward actionable fire risk mitigation planning
(See section on Project Objectives and Research Questions below).
1

The community's needs and interests have been prioritized above institutional goals
and directives by ensuring the community has had full ownership, control, and access to all
the data and by maintaining full transparency about the research process, methodologies, and
outcomes. These actions reflected the principles of “Ownership, Control, Access, and
Possession” (Absolon & Willett 2004; Absolon & Willett 2005; First Nations Information
Governance Centre, 2014). The Xáxli'p community owns all the data used in this project,
including the Airborne Laser Scanning (ALS) data, field data, and pre-existing data layers
from the Xáxli'p Community Forest (XCF) computer archive. They have full control over
how data is used, complete access to all information, and the authority to decide who has data
access. The research team ensured that all data were securely stored during the project.
Respecting these principles has been fundamental to building trust and ensuring the research
serves the community’s best interests.

Figure 2. View from the valley bottom looking into a farm field in the Fountain Valley.

2

Research Partnership Background and Context
This research partnership has been a collaborative co-creation process from the start,
led by the Xáxli’p Community Forest board of directors and management staff, with me
serving as the principal graduate researcher under their direction. The Xáxli’pmec People,
ancestral stewards of the Xáxli'p Survival Territory as the Xaxli’p refer to it, have
demonstrated their commitment by guiding this project. They set objectives to ensure the
outcomes align with their eco-cultural restoration priorities, addressing both immediate and
future needs (Diver, 2016b).
The Xáxli'p Survival Territory, also known as the Fountain Valley, is located just east
of Lillooet, BC, in the Pavilion Ranges eco-section along the Fraser River. Covering an area
of about 32,000 hectares, it is an ecologically sensitive area with steep topography,
dominated by dry-land Ponderosa Pine and Douglas Fir forests, with an arid climate due to
the rain shadow effect of the Coast Mountains (Xáxli'p Community Forest Corporation
[XCFC], 2009). The area faces a significant risk of catastrophic stand-replacing wildfires due
to historical fire suppression, the banning of traditional fire-keeping practices, and industrial
forest management (Bezzola, 2020).

3

Figure 3. Map illustrating the location of the study area within British Columbia, Canada, with the
main map highlighting the Xáxli’p Survival Territory in regional context and an inset on the left
providing a detailed outline of the Survival Territory boundary.

I joined the project in 2019 at the beginning of Phase Two, when a previous graduate
student, Aita Bezzola, was completing her master's thesis project with the Xáxli'p community
under the supervision of Dr. Scott Green (Phase One). Bezzola's work laid the foundation for
my involvement, identifying a critical concern related to land management: wildfire risk
throughout the Survival Territory and the lack of adequate, spatially specific information for
informed and effective risk-mitigation planning. Potential catastrophic wildfires in the
Xáxli'p Survival Territory pose an existential threat to the Xáxli'p way of life and “cultural
continuity” (Bezzola, 2020).
Our work follows a long history of knowledge co-production and intergovernmental
relations that the Xáxli'p community has undertaken dating back to the early 1900s (Diver,
4

2016). Bezzola's work in the Phase One project was built on decades of previous work by the
community and many allied researchers and practitioners, including postdoctoral researcher
Sybil Diver, forest ecologist Herb Hammond, and others. Their work highlighted significant
changes in the landscape and forest ecosystems throughout the Survival Territory in recent
decades, contrasting sharply with the ecological conditions that had been maintained by
Xáxli'p ancestral management practices for centuries prior (Diver, 2016; XCFC, 2009).

Figure 4. Aerial view of a forest plot surveyed during the 2020 field sampling campaign in the
northeastern section of the Xáxli'p Survival Territory. This image shows the higher density fuel types
found in the higher elevation areas of the Survival Territory and was captured by drone.

Bezzola’s work revealed the significant risk of high-intensity wildfire throughout the
Survival Territory, highlighting the potential for severe impacts on environmental conditions
and cultural values. She developed wildfire fuel and risk maps utilizing available data, which
5

identified well-connected, highly flammable fuel-loaded forests throughout the Survival
Territory. Coupled with the extreme topography and hot, dry climate of the territory, the
project identified a substantial risk of high-severity wildfire, something well-known by the
community but not formally reported and defined in detail until her research (Bezzola, 2020).
Key recommendations from Bezzola’s work included collecting Airborne Laser Scanning
data (a.k.a. LiDAR data) to improve forest fuel map resolution, managing grassland fire risk
to prevent spread into high-hazard forest areas, and continuing forest thinning, particularly
in high-risk zones. Additionally, designing fuel management areas to create fire-management
breaks and installing a weather station to better record local temperature and wind patterns
were advised (Bezzola, 2020).

Figure 5. Open field near Kwotlenemo Lake (Fountain Lake). Illustrating the recommendation from
Bezzola’s work regarding managing grasslands to avoid fire spread to surrounding high-hazard
forested areas.

To develop locally calibrated, higher-resolution maps of forest fuels, the community
procured funding for the acquisition of ALS data for a Phase Two project, where my
involvement began. Initially, using the ALS data to produce better maps of these fuel loads
6

became my key focus. During this work, a second key focus of the project emerged from the
co-development process: to provide a more effective way for the community to interact with,
access, and use the information generated from the fuel mapping of Phase Two, as well as
the pre-existing information and maps from previous work. Previously, maps and information
were primarily accessible in paper printouts and PDF files or were buried in complicated
folder structures and old geodatabases that required technical expertise and specific GIS
internal capacity that was not consistently available to the community. To address this need, a
user-friendly, interactive web-based mapping interface was developed to integrate Xáxli'p
maps and information from past work with the newly generated layers and field data, making
all of the information more accessible and available to support wildfire management, land
management, and ongoing climate-adaptation planning, while also supporting cultural
continuity and knowledge sharing. This approach enables knowledge mobilization and
allows the community to interact with the data more effectively.
This project evolution highlights the importance of adaptability as a core operating
principle for the research team. While the project had specific initial objectives, the ability to
respond to emerging community needs ensured that the outcomes remained relevant and
useful.
A key aspect of this project became comparing two different approaches to collecting
field data (described fully in Chapter 3). One approach followed a structured, scientific
method, using a formal quantitative sampling plan with precise measurements taken at
randomly selected sites across the landscape. This method, which is the “structured
approach,” follows established scientific procedures and allows for detailed, standardized
measurements. The second approach reflected local knowledge and direct observation, where

7

Xáxli'p community members selected locations based on their personal understanding of the
land and assessed forest conditions using their experience. This method, which can be
considered the “intuitive approach,” reflects community knowledge and practical expertise.
By integrating these two approaches, the project sought to bridge scientific methods with
local Indigenous knowledge,demonstrating how both forms of knowledge can complement
each other. This integrated approach also explored how modern technology can be adapted to
fit community-based decision-making rather than requiring communities to rely primarily on
external expertise.
Recent advancements in computer-based pattern recognition (machine learning)
present new opportunities to blend local and Indigenous knowledge with modern scientific
tools. Machine learning is a way for computers to recognize patterns in large amounts of
information. Although this method uses advanced technology, the principle behind it is
simple: it learns from experience, just like people do. In this project, a type of machine
learning called Random Forest modeling was used to identify patterns in the data and create
forest fuel maps. By training the model with both structured measurements and local
observations, the project ensured that the maps reflected Xáxli'p’s knowledge and priorities
rather than relying only on external scientific data.
Project Objectives and Research Questions
The objectives for this work were identified through a community-directed approach,
with the Xáxli'p Community Forest (XCF) Board of Directors serving as the primary
decision-making body. The board determined the project's overall priorities, objectives, and
approaches, guiding the direction of this research even before the Phase One project formally
began. The role of the research team has primarily been to function as technical support,
8

providing relevant and locally calibrated data for the community to advance its priorities,
most importantly, self-determination and cultural continuity.
The technical aspects of the project were developed in close collaboration with XCF
staff, the end-users of the data and mapping tools. Throughout the process, community
meetings, group sessions, and workshops were used to clarify objectives, provide updates on
progress, refine data needs, and offer training on how to use the information effectively. The
initial objectives were to map forest fuels in the Xáxli'p Territory using high-resolution
airborne scanning data (ALS), establish forest sampling plots for model validation, and use a
pattern recognition technique (called machine learning, described in Chapter 2) to analyze
the data. The objectives were adopted to ensure that the results would be usable, meaningful,
and accessible to the community in a non-technical format. By making this information
widely available, the project aimed to strengthen local knowledge-sharing and cultural
continuity, reflecting the community’s values and historical context.

Figure 6. The XCF crew and the UNBC research team discussing mapping and fieldwork planning
within the Xáxli'p Survival Territory.

9

The ultimate goal of this work was to ensure that local knowledge is integrated into
technical mapping and decision-making, not just as background information but also as the
foundation of the process. By using technology in a way that respects and reflects the Xáxli'p
relationship with the land, the project has helped create tools that are directly useful to the
community for wildfire management, land stewardship, and climate adaptation. To make this
knowledge truly accessible, a major focus was placed on making the maps and information
clear, intuitive, and easy to use, ensuring that anyone in the community, regardless of
technical background, can engage with the results. Guided by these objectives, the project
broadly sought to answer the following questions:
1. How can high-resolution airborne scanning data (ALS) help create detailed forest fuel
maps that are useful to the Xáxli'p community?
2. How can traditional/local knowledge be combined with scientific methods to improve
forest fuel mapping and wildfire management?
3. What are the differences, benefits, and challenges of structured scientific
measurements versus local knowledge-based observation in field data collection?
4. How can the maps and information from this project be made as useful and accessible
as possible to the community?
The Xáxli'p Survival Territory and Landscape Restoration History
The Xáxli'p are part of the St’at’imc Nation, a group of eleven Indigenous
communities that united in 1911 by signing the ‘Declaration of the Lillooet Tribe’ to assert
sovereignty and land rights against the BC provincial government's claims (Diver, 2016). As
of June 2024, the Xáxli'p community had a registered population of 1122, with about 259
individuals living on Xáxli'p reserve lands (Indigenous and Northern Affairs Canada, n.d.).
10

The Xáxli'p People have inhabited their Survival Territory along the Fraser River, in
the Fountain Valley, and the surrounding areas for thousands of years. Evidence of human
occupancy in the area dates back approximately 8,000 years, with archaeological sites
indicating village settlements as far back as 2,600 years (Prentiss & Kuijt, 2012). The word
“Xáxli'p” translates to “brow of the hill,” referring to a level spot at the mouth of the
Fountain Valley overlooking the Fraser River (Xáxli'p Government & 3PIKAS, 2024).
However, this is not just a geographical location; it is a relational vantage point from which
two distinct realities can be seen: "a small, lush place in an arid region", the Xáxli'p Survival
Territory, with its unique hydrological conditions,and the surrounding, drier landscape. Their
Survival Territory "has been a land of plenty when cared for" (Weinstein, 1995), highlighting
the deep contrast between these realities and the fundamental responsibility of the Xáxli'p
People to care for the land. This responsibility is at the heart of what it means to be Xáxli'p,
shaping their identity and ongoing stewardship of the territory.

11

Figure 7. View looking north over the south end of Kwotlenemo Lake in the central Fountain Valley
showing the high-density, connected fuels throughout the valley.

For centuries, the Xáxli'p people practiced fire-keeping and selective harvesting,
maintaining a healthy and resilient forest ecosystem. These practices supported numerous
plant and animal species and promoted a productive habitat for culturally significant flora
and fauna (Bezzola, 2020; XCFC, 2009). Consistent with many Indigenous cultures
worldwide, the Xáxli'p consider the lands and their inhabitants as part of their “family”
(Weinstein, 1995). However, institutional forest practices, such as clear-cut harvesting,
silviculture, and monoculture replanting, have disrupted this eco-cultural balance, leading to
dense, flammable forests (Diver, 2016; Moritz et al., 2014; Reinhardt et al., 2008).
Additionally, the imposition of total fire suppression and the banning of traditional firekeeping practices have exacerbated this issue, contributing to the current heightened wildfire

12

hazard and threatening degradation of Xáxli'p cultural continuity (XCFC, 2018), a concept
explored further in a subsequent section (Xáxli'p Cultural Continuity).
In response, over the past several decades the Xáxli'p community has undertaken
extensive work to safeguard their community and land-use priorities through fuel treatment,
thinning, and fire-smarting work, creating zones of lower fuel density to slow down fires
(Xáxli'p, 2022). They are acutely aware that, like their neighbors and much of the North
American continent, it is not a matter of if a wildfire comes but when and the intensity. The
changing climate, with more severe droughts, hotter temperatures, and longer fire seasons,
intensifies this threat (Wasserman & Mueller, 2023). A key land-use objective for Xáxli'p
people is to mitigate this emerging wildfire risk as much as possible so that when fires do
come, they will be less destructive and more manageable, posing less of an existential threat
(Xáxli'p, 2022). Figure 8 shows the sign displayed at the entrance to the Fountain Valley
community, indicating just how front of mind this concern is.

Figure 8. Entrance to the Xáxli'p First Nation Fountain Valley community. The Xáxli'p First Nation
sign reads: “Be Fire Smart Aware”.

13

Past projects in the Xáxli'p Survival Territory have focused on ecosystem restoration
and wildfire risk reduction. For example, the Diablo Meadows Project thinned a 13-hectare
forest from 650 trees per hectare (tph) to 400 tph to address overstocking caused by fire
suppression. Similarly, the Gibbs Creek Project reduced tree density from 400 tph to 200 tph.
These efforts aimed to enhance forest resilience to wildfire and restore ecosystems to their
pre-suppression state (Ubcwiki, 2021).
The core land management concerns for the Xáxli'p, in addition to wildfire risk,
include hydrological impacts, small space ecology conservation, road and private property
development in the valley, visual impacts, imposed resource management policies, and the
implications of these factors on community knowledge (Bezzola, 2020; Weinstein, 1995;
XCFC, 2009). Sustainable management for the Xáxli'p involves a “caring exchange,” taking
only what is needed and giving back to the land to ensure future growth. This is seen as
“culture in action” (Weinstein, 1995).
Modern challenges have disrupted balance throughout the territory. Industrial
logging, aggressive cattle grazing, and imposed land-management practices have degraded
the hydrological health of the Fountain Valley watershed and fragmented traditional trail
networks (Xáxli'p Community Forest, no.date). Private property development has also
reduced access to important berry-picking spots (Xáxli'p Community Forest, n.d.). The
Xáxli'p have always expressed that “without water, we are lost” (Weinstein, 1995).

14

Figure 9. A wildfire risk reduction fuel-treated area near Kwotlenemo Lake showing the treated area
(right) next to the untreated area (left), highlighting the notable difference in fuel loads.

Figure 10. An ALS generated forest fuel load map displayed on an iPad in the field within a fueltreated area of the Xáxli'p Survival Territory. The map indicates areas of high fuel density in red and
low density in yellow and green, with the blue dot representing the user’s location. The area on the
right shows a dramatic increase in understory and ladder fuel directly resulting from fire suppression.

15

The Xáxli'p Community Forest Agreement
As the BC government began to formalize crown lands throughout the province in the
early 1900s, the Xáxli'p signed the Declaration of the Lillooet Tribe in 1911 as an act of
sovereignty over their territory (Diver, 2016). Since then, the Xáxli'p have taken various
steps to regain control over their land and resources. Following the signing of the Declaration
of the Lillooet Tribes and the increase in forest harvesting throughout the province and within
the Xáxli'p Survival Territory, the community organized roadblocks in the 1990s to protest
commercial clearcut logging in the Survival Territory. In 1993, the Xáxli'p entered treaty
negotiations with the BC government but withdrew in 2001 due to frustrations with the
process (Diver, 2016).

Figure 11. Harvested areas in the high-elevation northwestern portion of the Xáxli'p Survival
Territory. These were replanted using both conventional and experimental silviculture techniques.

During the treaty negotiation period in the 1990s, the Xáxli'p conducted their own
traditional land-use planning process in collaboration with forest ecologist Herb Hammond,
16

beginning in 1997. This ecosystem-based planning framework gathered and organized
cultural and ecological data, drawing on existing information, to identify areas for protection
and sustainable forestry (Xáxli'p Community Forest, n.d.). The resulting community maps
and plans were used in policy negotiations, leading to the creation of the Xáxli'p Community
Forest (CFA K3L) in 2011 and, subsequently, the Xáxli'p Community Forest Corporation
(XCFC), which encompasses a large part of the Xáxli'p Survival Territory (Diver, 2016;
XCFC, 2009). See Figure 12 for a timeline of events and Figure 13 showing the boundary of
the Xáxli'p Community Forest.

Figure 12. Timeline of events leading up to the Xáxli’p Community Forest (XCF) Agreement
between the Xáxli’p community and the B.C. Ministry of Forests (MoF)” (Diver, 2016).

17

Figure 13. Map showing the boundary of the Xaxli’p Survival Territory (orange) overlaid on the
boundary of the Xaxli’p Community Forest (blue).

The Xáxli'p community chose this research partnership as a strategic path forward to
provide them with greater self-determination and flexibility to function within governmental
frameworks. This objective aligns with the trend observed by Goetze (2005) and Diver
(2012), where Indigenous communities participate more fully in environmental policy and
land management by establishing their own scientific programs or utilizing existing policy
frameworks to facilitate alternative and traditional practices (Diver, 2016).
By thoughtfully adopting locally calibrated data and technology from this project and
ensuring they align with their cultural values and long-term objectives, the Xáxli'p

18

community has positioned themselves to better influence environmental and management
policy within their territory to better protect their resources.
As Tall Bear (2013) highlighted, the decisions Indigenous peoples make when using
dominant scientific frameworks are "political acts" that advance Indigenous selfdetermination. This project presents a pathway by which traditional wisdom and modern
technology can coexist under the direction of Xáxli'p community leaders, ensuring that their
cultural values and long-term objectives are respected and upheld. This blend of methods
described in the subsequent chapters not only safeguards their territory and culture but also
serves as a model for broader applications for other Indigenous communities.
Xáxli'p Cultural Continuity
The ultimate goal of this work was to support the Xáxli'p community’s selfdetermination and cultural continuity, not just in terms of survival as a people, but in
ensuring the continuance of a cultural way of thinking and being in relationship with the
land. The project's focus has been directed by the community's priority to maintain selfgovernance over their Survival Territory and to ensure that land-management practices align
with traditional cultural values, including long-term sustainability goals. By prioritizing
community interests over institutional objectives, this research commits to producing results
that directly benefit the Xáxli'p People, fostering resilience and adaptability in the face of
emerging environmental challenges. This approach emphasizes empowering the Xáxli'p
community, increasing internal capacity, co-creating knowledge, and developing new tools to
help protect Xáxli'p lands and ways of life against the growing threats of wildfire and climate
change.

19

Figure 14. Xáxli'p Community Forest Crew Leader, exchanging a quiet moment with a horse while
journeying through an open field during the field data collection campaign.

The Xáxli'p believe that land-use decisions should consider whether proposed landuse objectives and activities would have been acceptable to the grandparents of today’s elders
and how current decisions will impact future Xáxli'p generations (Weinstein, 1995). Ideally,
land-use decision-making should foster cultural reproduction among Xáxli'p youth,
reinforcing faith, interest, and value in traditional knowledge. Concepts such as "don’t overdo-it" and "take only what’s needed" (Weinstein, 1995) resonate with ecological
understandings that have gained momentum in Western scientific discourse since the mid1990s (Diver, 2016).
For the Xáxli'p, as mentioned previously, land use is understood as a relationship
where the land and all its inhabitants are viewed as "family" (Weinstein, 1995). Traditional
land-management practices, developed over centuries, reflect the principle that "what
happens to one happens to others," recognizing the interconnectedness of all things (Xáxli'p
Community Forest, n.d.). This perspective is integral to the economic livelihood and cultural

20

identity of the Xáxli'p People. It underscores the need for decisions that ensure the well-being
of future generations of Xáxli'p family members (Weinstein, 1995).
One of the most tangible indicators of success in achieving the project’s initial targets,
supporting community objectives related to cultural continuity, can be seen in the recent
hiring of new Xáxli'p Community Forest (XCF) staff. Consisting of young Xáxli'p
community members who became responsible for overseeing the data and data management
program developed through this work, the new staff directly address the issue of cultural
continuity. This outcome reflects the priority of investing in community-led research and land
stewardship, ensuring that knowledge, capacity, and decision-making power remain within
the community. Establishing these roles not only strengthens the long-term management of
Xáxli'p lands but also reinforces the ongoing intergenerational transmission of land-based
knowledge, which is key to sustaining the Xáxli'p way of life.
Conclusion
These broad questions have shaped the study, guiding the research through its
different stages. Chapters 2 and 3 investigated specific aspects to inform these larger themes,
and Chapter 4 will synthesize the findings to revisit the core objectives.
At the heart of this research has been the active participation and leadership of the
Xáxli'p community. From securing funding for the ALS data to sharing invaluable local
knowledge, the Xáxli'p People have guided the research to align with their cultural values
and practical needs. Through continuous engagement, via discussions, field trips, and
community meetings, the collaboration has been grounded in mutual respect and
understanding. This partnership has ensured that the research outcomes remain relevant and
beneficial to the community.
21

By integrating Xáxli'p community knowledge with science-based methods, this
project highlights the synergy between local knowledge, science, and modern technology.
This approach not only benefits the Xáxli'p community but also serves as a model for other
Indigenous and community-led research initiatives. It exemplifies how community-based
knowledge can meaningfully guide scientific research, achieving scientifically sound and
culturally resonant outcomes.
The detailed forest fuel maps produced by this work support holistic wildfire
management strategies, helping to mitigate the risk of catastrophic wildfires in the Xáxli'p
territory and support eco-cultural restoration. These maps serve as important tools for climate
adaptation planning, enabling informed decision-making about land management in the face
of changing environmental conditions. By incorporating local knowledge, the research
supports the Xáxli'p community’s efforts to maintain their sovereignty in land management,
preserve their ecological heritage, build local capacity, and promote cultural continuance by
passing along knowledge and wisdom to the next generations. This project highlights the
importance of community-led research, adhering to the principles of Ownership, Control,
Access, and Possession (OCAP) outlined by the First Nations Information Governance
Center (The First Nations Information Governance Centre [FNIGC], n. d.). By respecting
these principles, the research ensures that the data and findings serve the community’s
interests, supporting their self-determination and long-term sustainability goals.
The following chapters of this thesis shift the focus to the technical analysis, machine
learning modeling and landscape-level map generation. Chapter 2 details the methodologies
used to create the high-resolution forest fuel maps, providing a comprehensive overview of
the initial random stratified, structured empirical (SE) field sampling approach and the

22

subsequent modeling and map generation process. Chapter 3 examines the second field data
collection approach, opportunistic ocular estimates (OE), and compares the two different
sampling methods, SE and OE. It analyzes the results generated from each method, providing
insights into their effectiveness and the implications for future research and land management
practices. Chapter 4 brings these elements together, reflecting on the entire project and its
broader significance. It revisits the community-directed approach explored in Chapter 1, the
technical analyses presented in Chapters 2 and 3, and considers how these findings contribute
to the ongoing journey of Xáxli'p land management. A key focus of this final chapter is the
development of the web-based mapping tool, which serves as both a culmination of the
project and a foundation for future applications. By making complex data more accessible
and useful for the community, this tool embodies the core objectives of the project,
supporting self-determination, cultural continuity, and locally driven land stewardship.

23

Chapter 2: Integrating Airborne Laser Scanning Data and Random Forest Modelling
for Wildfire Forest Fuel Mapping in the Xáxli’p Survival Territory
Introduction
Wildfires are an important ecological process, shaping forest ecosystems and
maintaining biodiversity worldwide (Bowman et al., 2020; Moritz et al., 2014; Pausas &
Keeley, 2019). In recent decades, the frequency and severity of wildfires have escalated due
to climate change and other anthropogenic factors (Abatzoglou & Williams, 2016; Coogan et
al., 2019; Doerr & Santín, 2016; Parisien et al., 2023; Westerling, 2016). This intensification
poses significant risks to communities, ecosystems, and economies (Szpakowski & Jensen,
2019). Addressing these risks requires a comprehensive understanding of forest fuel loading
and how variations in forest structure influence fire behavior (Cowman & Russell, 2021;
Heisig et al., 2022). This chapter focuses on how advanced remote sensing technologies can
improve the ability to characterize and manage forest fuels.
Historically, wildfire fuel mapping in Canada has relied on the Canadian Forest Fire
Danger Rating System (FBP System), specifically using Fire Behavior Prediction (FBP)
System fuel types to estimate fire behavior under certain fuel conditions and infer wildland
fire hazard (Baron et al., 2024; Forestry Canada Fire Danger Group, 1992; Stocks et al.,
1989). While this system provides a valuable framework, it primarily categorizes risk based
on broad forest types, resulting in coarse spatial resolution. This approach does not
adequately account for variations in fuel loading within the same forest type, potentially
overlooking critical differences that could influence wildland fire hazard at a more local scale
(Alexander & Cruz, 2013; Nadeau et al., 2005; Perrakis & Eade, 2015).
Forest fuels are commonly stratified into four vertical categories: ground,
surface/understory, ladder, and canopy fuels (Cruz & Alexander, 2014; Keane, 2015; Ottmar
24

et al., 2007). For this study, these categories are described as follows: Ground fuels primarily
consist of the duff layer, including decomposing organic matter (not measured in this study).
Understory fuels include the litter layer, mosses, lichens, dead woody debris, herbaceous
vegetation, and short to medium-height shrubs, which can sustain fire spread. Ladder fuels,
comprised of tall shrubs, understory trees, loose bark, and dead branches on tree boles, create
a pathway for fire to move from the surface to the canopy. Canopy fuels, including live and
dead foliage, twigs, and small branches in the overstory, can result in intense crown fires and
facilitate widespread wind- and topography-driven fire growth (Cruz et al., 2003; González‐
Ferreiro et al., 2017). Separating and measuring these fuel strata is very valuable for
identifying areas of varying fire risk and enabling targeted fire management interventions
(Agee & Skinner, 2005; Coogan et al., 2019).
Airborne Laser Scanning (ALS), or Light Detection and Ranging (LiDAR), has
emerged as a transformative technology for characterizing forest structures in three
dimensions with high precision (Andersen et al., 2005; Gajardo et al., 2014; White et al.,
2013). ALS provides detailed information on the vertical and horizontal distribution of
vegetation, offering the potential to directly quantify different fuel layers. Recent studies
have demonstrated the effectiveness of ALS data in deriving forest structure information for
fire danger modeling and assessing fire severity (Hollaus et al., 2021; Montealegre et al.,
2014). However, questions remain about how best to utilize ALS data to inform fire risk
assessments and integrate it effectively with existing wildfire management systems
(Fernández-Álvarez et al., 2019; Fernández-Guisuraga et al., 2023; Hollaus et al., 2021;
Martin-Ducup et al., 2025).

25

The Xáxli’p Survival Territory, located in a fire-prone region of British Columbia,
provides a compelling case study for addressing these questions. The territory is located in a
part of British Columbia where historical and contemporary land management practices, such
as clear-cut harvesting and fire suppression, have significantly influenced the forest's
structure and fuel loading (Gayton, 2015; Klenner & Arsenault, 2008; Marcoux et al., 2021).
Additionally, discrepancies between on-the-ground observations and existing provincial and
national fuel maps underscore the need for improved mapping methods tailored to the region.
Recent studies by Phelps and Beverly (2024), and Baron et al. (2024) found poor
correspondence between field assessment data and provincial and national fuel types in
interior British Columbia, with mismatches particularly frequent for dry interior ecosystems,
mixed wood and deciduous fuel types, and post-harvesting conditions.
This study aims to enhance wildfire risk assessment by integrating ALS data with
Random Forest modeling, a machine-learning approach known for handling complex, highdimensional datasets (Breiman, 2001; Cutler et al., 2007). By combining advanced remote
sensing technologies with local community and Indigenous knowledge, we seek to produce
detailed and relevant forest fuel maps that improve conventional methods that have produced
the currently available fuel maps. This project has been fundamentally community directed
from its inception, with the principles of Ownership, Control, Access, and Possession
(OCAP) guiding the collaborative approach. These principles ensure that the research
respects the rights and interests of the Xáxli’p community (First Nations Information
Governance Centre, 2014; Schnarch, 2004). Under the guidance of the Xáxli’p Community
Forest Board of Directors (XCF BoD) and XCF management staff, all data are owned by the
XCF and have been securely stored by the research team throughout the project.

26

Specifically, this study addresses the following research questions:
1. How can we accurately evaluate the vertical and horizontal spatial distribution of
forest fuel components throughout the Xáxli’p Survival Territory using ALS data and
Random Forest modeling?
2. To what extent do fuel loading components in the study area vary within existing Fire
Behaviour Prediction (FBP) System fuel types, and what are the implications of this
variation for wildfire risk assessment?
3. How can ALS data and machine learning techniques improve wildfire fuel
classification beyond traditional fuel type categorizations, providing a more detailed
and accurate assessment of wildfire hazards?
By addressing these questions, this chapter contributes to the growing body of
knowledge on forest fuel mapping and provides a practical, community-informed approach to
wildfire risk management. Integrating advanced technologies with community knowledge
has the potential to enhance the accuracy, accessibility, and utility of forest fuel maps,
ultimately supporting more effective and sustainable wildfire management strategies in the
Xáxli’p Survival Territory and beyond.
Methods
This study employs a multi-step methodology to develop and validate models for
predicting wildfire forest fuel characteristics in interior southwestern British Columbia. The
approach integrates field sampling, Airborne Laser Scanning (ALS) data acquisition, and
Random Forest (RF) machine learning techniques.

27

Study Area
As mentioned in Chapter 1, the study was conducted within the Survival Territory of
the Xáxli’p First Nation, encompassing the Xáxli’p community forest tenure area (50°44' N,
121°51' W). This region, located east of Lillooet, British Columbia, in the Pavilion Ranges
eco-section, is also known as the Fountain Valley and covers approximately 35,000 hectares
(Demarchi, 2011; Forest Analysis and Inventory Branch, 2022).
The climate of the region is characterized by a mean annual temperature range of
5.4°C to 15.7°C, with average monthly temperatures ranging from -1.1°C in January to
22.6°C in July. The mean annual precipitation is 418.9 mm, with the wettest months being
November (66.8 mm) and January (64.3 mm), and the driest months being April (18.3 mm)
and August (17.8 mm) (Government of Canada, 2023).

Figure 15. Map illustrating the location of the study area within British Columbia, Canada. The main
map highlights the Xáxli’p Survival Territory, within the context of British Columbia. The inset on
the left provides an outline of the Survival Territory boundary.

28

The topography of the Fountain Valley is characterized by extreme variations, with
elevations ranging from approximately 200 meters above sea level along the Fraser River to
about 2300 meters at the highest mountain peaks. This diverse elevation gradient influences
local climatic and ecological conditions, creating a range of microclimates and habitats that
contribute to the area's biodiversity (Pojar et al., 1987). The extreme topography also presents
challenges for fuel management and increases the risk of slope and valley wind-driven
wildfire behavior (Rothermel, 1983).
The territory predominantly falls within the Interior Douglas Fir (IDF) and Montane
Spruce (MS) biogeoclimatic zones, characterized by dry and warm conditions (Pojar et al.,
1987). The conditions typical of these zones play a significant role in determining fire
behavior and fuel characteristics within the territory. The predominant tree species are
Douglas-Fir (Pseudotsuga menziesii) and Ponderosa Pine (Pinus ponderosa). Forest fuel
composition varies across the elevation gradient, with open dry Douglas-Fir and Ponderosa
Pine forests at lower elevations transitioning to the Montane Spruce zone with Engelmann
Spruce (Picea engelmannii) and Lodgepole Pine (Pinus contorta) stands at higher elevations
(Demarchi, 2011; Forest Analysis and Inventory Branch, 2022). Understanding this variation
in forest fuel types is essential for modeling and mapping forest fuels and managing wildfire
risks (Keane et al., 2001).
Douglas-fir and Ponderosa Pine areas exhibit open canopies and lower stem densities,
influencing ladder fuel distribution. In contrast, spruce forests have denser canopies and
higher moisture content, contributing to greater stem densities. Lodgepole Pine stands
display high stem densities with significant accumulations of ladder and crown fuels. The
Xáxli’p community identified key forest types of interest within their territory, highlighting

29

variations in structure and fuel characteristics. The community described "dog-haired"
stands, areas characterized by very dense young Douglas-fir beneath mature canopies, as
being of particular interest as they contribute to elevated understory and ladder fuels and
impact plant and animal habitats. However, due to operational feasibility and time
constraints, field data collected excluded smaller trees whose diameter at breast height
(DBH) was less than 10 cm, potentially underestimating understory stem density in these
areas.
Historically, the Xáxli’p employed fire-keeping practices and selective harvesting to
maintain balance in the Survival Territory (Bezzola, 2020; Diver, 2016). These practices
contributed to the preservation of important animal habitats and promoted the growth of
culturally significant plants (Turner et al., 2000). The community's traditional fire practices
and selective harvesting helped maintain the health and resilience of the forest ecosystems by
ensuring a diverse mosaic of different forest structures, stand ages, species compositions, and
fuel densities throughout the territory. This diversity enhanced the forest's resilience to
catastrophic stand-replacing fires (Christianson, 2015; Diver, 2016).
In contrast, modern institutional forest management practices, such as clear-cut forest
harvesting and silviculture practices, predominantly result in single-age, high-density
monoculture stands (Puettmann et al., 2009). Moreover, decades of aggressive forest fire
suppression have altered the natural fire regimes and ecosystem dynamics, leading to
increased fuel loading and vulnerability to large, severe wildfires (Hessburg et al., 2005;
Puettmann et al., 2009). Understanding these historical and contemporary management
practices and their impacts is crucial for developing effective fire management strategies that
restore balance and resilience to the forest ecosystem (Odion et al., 2014).

30

The surrounding communities engage in small-scale timber harvesting and
agricultural activities, contributing to the local economy (Forest Analysis and Inventory
Branch, 2022). Several areas within the territory are undergoing fuel treatments and
ecosystem restoration initiatives aimed at enhancing fire resilience around populated and
critical habitat areas (XCFC, n.d., 2009, 2018). These community and economic activities are
important to the overall management of the land and its resources, emphasizing the need for
sustainable practices that balance economic benefits with ecological health (Charnley et al.,
2007).
Airborne Laser Scanning (ALS) Data
The Airborne Laser Scanning (ALS) data was collected using a Riegl VQ-580
airborne laser scanner mounted on a Haikai Research Institute fixed-wing aircraft. This
equipment achieved a high-resolution scan of approximately 13 points per square meter. The
complete dataset was flown over the entire Xáxli’p Survival Territory during the summer of
2020, providing wall-to-wall coverage (Riegl Laser Measurement Systems GmbH, 2018).
The ALS, also known as Light Detection and Ranging (LiDAR), is highly effective for
predicting detailed forest biophysical structures and inventory variables at various scales.
This technology provides accurate and timely information about forest structural
characteristics, which is critical for effective wildfire management. Previous studies have
demonstrated the utility of ALS in capturing detailed forest attributes such as canopy
structure and biomass distribution (Ahmed et al., 2015; Mutlu et al., 2008; White et al., 2013;
Woods et al., 2011).
Numerous forest metrics were derived from the ALS point cloud data, including tree
height, stand height, basal area, and volume. These metrics are important for developing
31

forest fuel load maps. Key metrics such as canopy bulk density, canopy base height, canopy
fuel loading, stems per hectare, ladder fuel load, and understory fuel load provide valuable
information for evaluating and predicting wildfire behavior (Ahmed et al., 2015; Chen et al.,
2016; Coops et al., 2016; Pimont et al., 2016; Zhao et al., 2011).
The high-density data enables comprehensive mapping of the area's forest structure,
capturing detailed information essential for accurate forest fuel mapping. However, while
ALS is highly effective for accurately mapping above-ground forest fuel structural attributes,
its capability to detect and measure the ground fuel layer is still under-researched (Bright et
al., 2017). This study focused on above-ground forest fuel structures, acknowledging the
significant role of ground fuels in fire dynamics but not including them in the detailed
mapping efforts. The limitations of ALS in capturing ground fuel data highlight the need for
complementary methods to achieve a comprehensive understanding of forest fuel dynamics.
Field Sampling Framework and Campaign
To ensure a comprehensive representation of the diverse forest types and landscape
attributes across the entire study area, we employed the Conditional Latin Hypercube
Sampling (cLHS) sampling approach. Developed by Minasny and McBratney (2006), cLHS
optimizes sample locations using a combination of continuous and categorical covariates,
making it particularly effective in ecological studies with high spatial heterogeneity. The
cLHS method is a random stratified procedure that selects sampling locations based on prior
information on a suite of environmental variables. This approach has been widely used in
digital soil mapping projects and other ecological studies worldwide, as it effectively
captures the variability of environmental factors. It reduces selection bias and captures a wide

32

range of environmental conditions, ensuring that the sample distribution is both statistically
efficient and environmentally representative (Roudier et al., 2012).
Our sampling strategy incorporated layers of Fire Behaviour Prediction (FBP) System
fuel types, elevation data, and accessibility constraints. The 2019 fuel-type layers were
provided by the British Columbia Wildfire Service (BCWS) through a data-sharing
arrangement established in 2019 before the BCWS, making the fuel-type layers publicly
available. Elevation data was used to capture landscape variations influencing vegetation
types, microclimates, moisture regimes, and the general environmental variability across the
territory. At the time of sample design, we did not have access to landscape-level ALSderived products to incorporate into the stratification. The FBP System fuel types, derived
from the Vegetation Resources Inventory (VRI) and other government data, provided a
practical framework despite their coarse resolution. This pragmatic decision allowed us to
advance the project efficiently with a robust sampling framework (Stocks et al., 1989).
Community engagement was integral to our sampling strategy. We consulted
extensively with the Xáxli’p community to gather insights into road conditions, vehicle
accessibility, and local ecological knowledge. This input was crucial for accurately
classifying roads, updating the comprehensive road layer, and refining sampling locations.
The community's contributions were instrumental in identifying deactivated roads and areas
to avoid, providing a nuanced understanding of the landscape and accessibility
considerations.
To ensure accessibility and safety during field sampling, we implemented a 200-meter
buffer around all accessible roads, restricting the selection of sample points within this
distance. This approach accounted for practical field conditions and logistical constraints,

33

ensuring that field crews could safely access all sample sites efficiently. With these layers, we
created distinct strata based on elevation intervals, fuel types, and proximity to forest service
roads. Within each stratified area, the cLHS method was used to randomly select potential
sample sites, improving statistical efficiency by ensuring that each site represented the
variability within its stratum.
Originally, the sampling plan targeted 110 sample sites, but instances of inaccessible
terrain and ecological obstacles encountered in the field necessitated some adjustments,
reducing the number to 104. These changes were essential for field crew safety and site
feasibility. Flexibility and adaptability in fieldwork allowed us to respond to on-the-ground
realities, ensuring that data collected accurately reflected current conditions. To account for
missed plots, we collected approximately 5-10 additional samples from areas that had
undergone recent treatment or showed significant ecological changes not initially accounted
for in the cLHS design, contributing to the 104 total sample sites. These additional samples
were valuable for ensuring comprehensive coverage and robust data for reliable analysis and
conclusions.
Safety protocols were paramount throughout the sampling campaign. Protocols
included regular communication, emergency preparedness plans, and accessibility to first aid
resources, ensuring the field crew could respond quickly to any potential hazards in the
rugged terrain. Adjustments were meticulously documented to provide context for the data
analysis phase, meaning any potential biases introduced by these changes were understood
and accounted for. Adjustments made during the field campaign reflect the flexibility
required in ecological research, demonstrating how methodological rigor and responsiveness
to field conditions can enhance research outcomes (Roudier et al., 2012).

34

Figure 16. Map depicting the extent of the Xaxlip Survival Territory study area. Blue dots represent
the distribution of the 104 sample locations across the territory. Sample locations were restricted to
within 200 meters of roads, which influenced their distribution throughout the territory.

35

Techniques, Equipment, and Data Collection
To gather empirical field data for training and calibrating the fuel models, we
collected a range of measurements, including tree diameter at breast height (DBH), total
height, crown base height, species identification, and canopy density assessments of both live
and dead trees. These methods captured forest attributes crucial for fuel load and forest
structure modeling. Field teams used Differential GPS (DGPS) systems for precise
georeferencing of plot locations, minimizing errors. Other essential tools included DBH tape
measures, hypsometers, clinometers, iPads with EpiCollect for data recording, and flagging
tape to mark plot boundaries and features. Sample plots of 400m² were chosen to balance
detailed data capture with the intended 100-m² grid cell size for analysis. This plot size
helped to minimize edge effects and achieve appropriate detail representative of the
horizontal diversity of the local forest types being studied. Plot centers were occasionally
adjusted to avoid edge effects or varying vegetation types and densities, maintaining data
quality and relevance.

36

Figure 17. Diagram of forest plot layout showing a large 11.28m radius plot with four smaller 2.99m
radius sub-plots positioned at 0, 90, 180, and 270 degrees.

Two types of field plots were utilized for data collection: large plots with an 11.28 m
radius (400 m²) and small plots with a 2.99 m radius (28 m²); see Figure 17. Within the large
plots (11.28 m radius, 400 m²), all trees with a DBH greater than 10 cm were measured. For
each tree, the following parameters were recorded: DBH, total height, height to the base of
the live crown, height to the base of the dead crown, species, crown position (classified as
dominant, intermediate, co-dominant, or understory), crown class (percentage of the canopy
circumference that is complete/filled out), and decay class (ranging from 1 to 9).

37

In the small plots (2.99 m radius, 28 m²), visual estimates were made for the density
of different fuel layers. These layers were categorized as understory (0–1.37 m,
approximately breast height), ladder fuel (1.37–3 m, approximately average canopy base
height), and canopy fuel (3 m and above, from the canopy base to tree tops). Density values
for these layers were recorded using a scale of 0–25% (low density), 25–50% (low/medium
density), 50–75% (high/medium density), and 75–100% (high density). Data collection
protocols were developed in consultation with BC Wildfire Service experts and informed by
best practices from forestry research literature, ensuring robust and reliable data suitable for
predictive modeling of forest dynamics (Andersen et al., 2005; White et al., 2013).
Data Processing
The ALS data underwent quality assurance using the lidR package in R (Roussel et
al., 2020), adhering to industry standards for ALS data processing, including applying
elevation cut-offs and standard classification practices. A Digital Elevation Model (DEM)
was generated using the Triangulated Irregular Network (TIN) technique with standard
settings from the lidR package. This DEM served as the baseline for normalizing the ALS
data, ensuring accurate above-ground measurements of tree and vegetation heights (Roussel
et al., 2020). From the normalized point cloud, we calculated a comprehensive set of standard
ALS metrics, along with custom metrics developed specifically for this study.
By correlating field-based data with ALS data, area-based approaches (ABA)
techniques were utilized to provide stand averages for relatively homogeneous forested areas.
The analysis was performed on a 10 m raster cell size scale, balancing the desired highresolution output with the field sample plot size (11.28m diameter circles). This resolution
effectively describes differences in fuel distributions throughout the Xáxli’p Territory,
38

yielding measures such as stand volume, basal area, mean heights, and diameters. These
ABA techniques have been extensively used to derive forest metrics, such as maximum
height, mean height, and height percentiles (van Ewijk et al., 2011; Vastaranta et al., 2012;
White et al., 2013). These metrics, crucial for subsequent modeling, were computed using the
lidR package in combination with custom scripts that incorporated field data, allometric
calculations, ALS height bins, and point densities to derive complex forest structure attributes
like Canopy Bulk Density (CBD), Crown Fuel Load (CFL), Canopy Base Height and several
others.
Using allometric equations from Ung et al. (2008), field measurements were
transformed into the key forestry metrics, and then the derived metrics were aligned with
ALS-derived metrics to create a unified dataset. This dataset served as the training data for
subsequent Random Forest modeling, providing a comprehensive view of the forest's
structural attributes. Before modeling, the dataset underwent cleaning, including checks for
data integrity, to ensure robust model performance (Jenkins et al., 2004; Liaw & Wiener,
2002).
Three critical vertically distinct fuel layers were identified by the community of
particular interest, understory, ladder, and canopy fuels. The understory fuel layer
encompassed vegetation from ground level to 2 meters, reflecting surface-level fuel loads.
The ladder fuel layer, crucial for understanding vertical fire spread, included vegetation from
2 meters to the median canopy base height. The canopy fuel layer represented fuels from the
median canopy base height to the uppermost parts of the canopy. A custom binning function
was applied to airborne laser scanning (ALS) data to group point densities within these
height intervals. For detailed definitions of these metrics and variables, refer to Appendix 1.

39

This study used standard metrics generated through the LidR package and custom
metrics developed specifically for this analysis to calculate important landscape-level fuel
variables of interest. Standard metrics include commonly derived variables such as maximum
vegetation height, mean vegetation height, and vegetation density at specific height intervals.
Custom metrics focused on measures like canopy base height, canopy density, and vertical
foliage distribution.
Random Forest Modeling
This study employs Random Forest modeling to predict key forest fuel attributes
across the Xáxli’p Survival Territory using Airborne Laser Scanning (ALS) data. To achieve
this, ALS-derived structural metrics serve as independent (explanatory) variables, while fuelrelated attributes are modeled as dependent (predicted) variables. The model leverages the
Random Forest algorithm, an ensemble machine learning approach, to capture complex, nonlinear relationships between forest structure and fuel loading while minimizing overfitting.
A summary of key model variables is provided below, with a comprehensive list and
definitions available in Appendix 2.

40

Table 1. Summary of Model Variables

Category

Variable Type

Description
Measures of canopy
height, density, foliage
Independent Variables
distribution, and intensity
(Predictors)
ALS-Derived Metrics
from ALS.
Leaf area density, canopy
base height, biomass
Structural & Fuel Metrics
estimates.
Dependent Variables
Fuel available in the
(Predicted Outputs)
Canopy Fuel Load (CFL)
canopy layer.
Fuel that facilitates
Ladder Fuel Load (lfuel)
vertical fire spread.
Low-level fuels
contributing to fire spread
Understory Fuel Load
from surface to ladder
Understory Fuel (ufuel)
(ufuel)
fuels.
Canopy Bulk Density
(CBD), Median Canopy
Base Height (Med.CBH),
Stems per Hectare,
Additional Forest Attributes Biomass Estimates.
For a full breakdown of variables used in model training and prediction, see Appendix 2.
The Random Forest technique was applied using a comprehensive modeling
approach, incorporating all available predictor variables to maximize the model's potential to
identify significant patterns and relationships. This approach leverages Random Forest's
ability to handle high-dimensional data effectively while minimizing the risk of overfitting.
Variables prone to errors or lacking complete coverage for landscape-scale prediction were
excluded to ensure model reliability and computational efficiency (Biau & Scornet, 2016;
Breiman, 2001).
The Random Forest (RF) method is an effective tool for analyzing ecological data,
especially for mapping forest fuels. It combines predictions from many decision trees, each
trained on different parts of the data, to improve accuracy and reduce errors. A decision tree
is a flowchart-like model where each node represents a decision based on a feature or metric
41

(like tree height), each branch represents an outcome, and each leaf node represents a final
prediction. The RF method handles complex data without needing to pre-select specific
features (e.g., tree height, tree diameter, canopy density, and species type), effectively
capturing diverse forest characteristics (Breiman, 2001).
The RF was chosen to model forest attributes for several reasons. First, the model's
ensemble nature reduces the risk of overfitting, a common issue in simpler tree models
(Breiman, 2001). Additionally, RF excels at capturing complex, non-linear relationships
inherent in natural phenomena without requiring assumptions about data distribution. This
capability is crucial when dealing with diverse forest attributes, such as fuel load
characteristics (Cutler et al., 2007). The RF model can rank the importance of different
variables' ability to predict the intended output, which is particularly valuable in this type of
study, where understanding the impact of various forest attributes on fuel load is useful.
Model performance was assessed using three key metrics: out-of-bag R-squared
(OOB R²), cross-validated R-squared (CV R²), and validation R-squared (Validation R²).
Out-of-bag R-squared is an internal validation metric specific to random forests that
estimates predictive accuracy using data not included in the bootstrap sample for each tree,
providing an unbiased evaluation of the model's performance without needing a separate
validation set. Cross-validated R-squared is obtained through 5-fold cross-validation, where
the dataset is divided into five subsets, and the model is trained and tested five times, each
time using a different subset for testing. This metric ensures robust assessment by averaging
performance across multiple train-test splits. Validation R² measures the model's predictive
power on an independent validation dataset that was not used during training, offering an
external validation of the model's generalizability to unseen data. Together, these metrics

42

comprehensively assess the model's accuracy, consistency, and ability to generalize across
different forest fuel strata and attributes.
The models were configured with key parameters optimized for performance: the
number of variables considered from every bagging sample (mtry) was set to 15, the number
of trees in the forest (trees) was set to 1000, and the minimum number of data points in a
node before it is split (min_n) was set to 3. The data was initially split into training and
testing sets, with 90% of the data used for training and 10% for testing, and a set seed of 321
to ensure reproducibility. Additionally, a threshold of 0.95 was used to remove highly
correlated predictors to reduce multicollinearity.
The RF modelling identified various ALS-derived metrics as the most influential
predictor variables for the different forest fuel layers that were predicted using the models.
These metrics, depicted in bar charts located (Appendix 3), highlight the relative importance
and roles of certain variables in the model's prediction process. While these detailed
visualizations may interest remote sensing experts, no significant patterns or insights were
gained from the emerging variables. Therefore, they may be difficult to interpret and less
meaningful for the broader audience of this thesis.
Following model training and validation, the predict() function in R was used to
generate the final landscape-scale predicted fuel layers across the entire Xáxli’p Survival
Territory. The trained RF models were applied to the complete set of ALS-derived predictor
variables, which were processed using the lidR package (Roussel et al., 2020). These
predictor variables, including standard and custom metrics derived from the ALS point cloud,
captured key forest structural attributes. Using the trained models, fuel values were predicted
across the entire study area, generating spatially explicit estimates of canopy fuel, ladder fuel,

43

and understory fuel, as well as additional forest attributes such as canopy bulk density and
median canopy base height. The resulting fuel layers were then normalized and symbolized
using an intuitive color scale to facilitate visual interpretation and comparison across the
landscape. These final predicted outputs are the foundation for subsequent correlation
analysis, aggregation, and difference mapping, providing a comprehensive view of fuel
distribution across the Xáxli’p Survival Territory.
Correlation Analysis of Model Outputs
We conducted a correlation analysis to assess the relationships between the forest
metrics predicted by the Random Forest model. FBP System defines Canada’s diverse forest
types into 16 distinct fuel types, requiring broad generalizations (Canadian Wildland Fire
Information System, n.d.). When specific forest types do not align with textbook
descriptions, the closest match is selected based on anticipated fire behavior, which can lead
to assumptions and oversimplifications. These generalizations can be practical for informing
operational firefighting response decision-making but may affect the accuracy of fire
management strategies due to a lack of resolution and detail. For instance, areas classified as
C-2 (Boreal Spruce) in this study may exhibit significant variability that is not fully captured
by this FBP System fuel type classification.
Key fuel types and their characteristics include C-2 (Boreal Spruce), with pure black
spruce stands, feathermoss ground cover, and Labrador tea as the dominant shrub; C-3
(Mature Jack or Lodgepole Pine), characterized by dense pine stands and light dead surface
fuels; C-5 (Red and White Pine), which features moderate understory and a ground cover of
pine litter; and C-7 (Ponderosa Pine/Douglas-Fir), with open canopies and grassy ground
cover. Others include D-1/2 (Leafless/Leafed Aspen), where fire-carrying surface fuels
44

include deciduous litter and herbaceous material; M-1/2 (Mixedwood), mixed stands where
fire spread depends on the softwood-to-hardwood ratio; and O-1a/b (Grass), which varies
based on the proportion of cured (dried) material (Wotton, 2008; Canadian Wildland Fire
Information System, n.d.).
This analysis focused on the three community-identified fuel layers, crown fuel,
ladder fuel, and understory fuel, in comparison to the conventionally and operationally used
FBP System fuel types to provide insights into their collective impact on forest fuel loads and
fire behavior across the landscape. The calculation process involved several steps to analyze
the relationships between the fuel layers. First, each fuel layer, crown, ladder, and understory,
was normalized to a scale of 0 to 1 to standardize the data and enable meaningful
comparisons. Stratified random sampling was then applied to select 1,000 representative
sample points within each fuel type. Fuel values for the crown, ladder, and understory layers
were extracted from the normalized raster layers at these sample points. The extracted data
was subsequently grouped by fuel type, creating subsets of data for each fuel type that
contained the corresponding fuel values.
To assess relationships, the Pearson correlation coefficient was calculated for each
pair of fuel layers within each fuel type, providing a measure of the strength and direction of
linear relationships. Finally, scatter plots with regression lines and box plots were generated
to visualize these relationships, with correlation coefficients annotated on the plots .
Furthermore, other important variables were considered in the analysis, although they
were not included in the discussion or results section of this chapter. See Appendix 2 for
more details. Correlation coefficients were calculated to determine the magnitude and
direction of relationships, and correlation plots were generated to visualize these

45

relationships. This facilitated the identification of correlations that warranted further
investigation or confirmation.
Aggregate Fuels and Difference Maps
The complete landscape-scale final fuel layers generated through the random forest
modelling process were symbolized using an intuitive color scale to indicate areas of higher
and lower fuel density. These layers include the normalized canopy fuel, ladder fuel, and
understory fuel, representing the three community-identified vertical fuel strata of interest, as
well as additional attributes such as stems per hectare, median canopy base height, and
normalized canopy bulk density. The maps provide a foundation for understanding the spatial
distribution of wildfire fuels across the study area and serve as a baseline for subsequent
analyses. They also inform whether emergent clusters of fuel loading classes, derived from
ALS data and machine learning techniques, provide a more nuanced and accurate
classification of fire risk across the landscape compared to traditional fuel type
categorizations.
Following the generation of these primary fuel layers, aggregate fuel layers were
calculated by normalizing fuel attributes and deriving unweighted and weighted aggregates
to represent different combinations of key forest variables. The normalized attributes
included understory fuel (ufuel), ladder fuel (lfuel), canopy fuel (cfuel), canopy bulk density
(CBD), median canopy base height (CBH), and stems per hectare (stemsperha). For the
unweighted aggregation, the normalized raster layers were simply added together to create
three aggregate layers. The first aggregate (agg0) included cfuel, lfuel, and ufuel. The second
aggregate (agg1) incorporated CBD and CBH alongside lfuel and ufuel, while the third
aggregate (agg2) extended agg1 by adding stemsperha.
46

Weighted aggregate layers were calculated to account for the relative importance of
each attribute, as determined by cross-validation R² values derived from model performance
metrics. Weighted aggregate 1 (agg1) combined CBD, CBH, lfuel, and ufuel, each scaled by
their respective R² values. Weighted aggregate 2 (agg2) expanded on this by including
stemsperha, also weighted by its R² value. Table 1 shows the cross-validation R² values used
for weighting each layer. By incorporating these weights, the weighted aggregates accounted
for the confidence and predictive strength of each fuel attribute, influencing the aggregate
layer output accordingly. The effect of weighting on the aggregate layers was analyzed by
calculating the differences between weighted and unweighted layers for both agg1 and agg2.
These differences were visualized through mapping to highlight variations introduced by the
weighting process and provide insights into the relative significance of each attribute within
the overall fuel layer framework.

Table 2. Cross-validation R² values used for weighted aggregation

Variable

CV_R²

Stems per Hectare (stemsperha)
Median Canopy Base Height
(Med_CBH)

0.777

Canopy Bulk Density (CBD)

0.680

Ladder Fuel (lfuel)

0.559

Understory Fuel (ufuel)

0.292

0.707

Additionally, difference maps were created to assess the variations between each of
the three community-identified fuel layers of interest: understory, ladder, and canopy. These
maps were generated by subtracting the normalized rasters for each pair of layers, resulting
in three different maps: canopy minus ladder, canopy minus understory, and ladder minus
understory. To enhance interpretability, an intuitive color scale symbology was applied,
47

highlighting areas of positive and negative differences between the layers. This visualization
provided a clear spatial representation of the variations between fuel layers, facilitating a
better understanding of their relationships and distributions across the landscape.
Results
Key Fuel Maps
The final fuel layers generated in this study provide a spatially explicit representation
of wildfire fuel distribution across the Xáxli’p Survival Territory. These layers illustrate the
distribution of canopy, ladder, and understory fuel, which represent the three communityidentified vertical strata of interest. Understanding the spatial arrangement of these fuel
layers is critical for assessing fuel structure and landscape heterogeneity. The primary fuel
layer maps depict the spatial variability in fuel density, showing areas of high, medium, and
low fuel loads across the landscape. These maps have been normalized to allow for direct
comparison and symbolized using an intuitive color scale
Figure 18 presents a side-by-side display of the three key predicted fuel layers,
canopy, ladder, and understory fuel, which form the primary focus of this analysis. In
addition to these primary fuel layers, several key attributes relevant to fire behavior modeling
were also derived and incorporated into further analyses. The additional key attributes
include stems per hectare, median canopy base height, and canopy bulk density, contributing
to a more detailed assessment of forest structure and fuel continuity. Figure 19 displays these
additional predicted fuel layers, which were included in the aggregate fuel layer calculations
to further refine the assessment of wildfire hazard.

48

Figure 18. A side-by-side display of the final output of the three key predicted fuel layers that have been the focus of this analysis.

49

Figure 19. A side-by-side display of the final output of additional predicted fuel layers that have been included in the aggregate fuel layers calculation.

50

Random Forest Model Performance
The performance of the models (Table 2) across different forest fuel strata,
understory, ladder, and canopy fuel, as well as additional forest attributes (Canopy Base
Height [CBH], Canopy Bulk Density [CBD], Canopy Fuel Load [CFL], and Stems Per
Hectare [stemspha]) was assessed using three key metrics: out-of-bag R-squared (OOB R²),
cross-validated R-squared (CV R²), and validation R-squared (Validation R²). These metrics
provide insights into the model's predictive accuracy and generalizability.

Table 3. Summary of performance metrics for each forest fuel type and additional forest attributes.

Response Variable
Understory Fuel
Ladder Fuel
Canopy Fuel
Median Canopy Base
Height (CBH)
Canopy Bulk Density
(CBD)
Crown Fuel Load
(CFL)
Stems Per Hectare
(stemsperha)

Cross-Validated (CV)
Out Of Bag (OOB) R² R²
0.159
0.292
0.501
0.559
0.784
0.752

Validation R²
0.534
0.265
0.713

0.653

0.707

0.895

0.659

0.680

0.742

0.796

0.770

0.690

0.696

0.777

0.596

Correlation Between ALS Generated Fuels and FBP Fuel Types.
Understanding the correlation between various fuel types provides valuable insights
into the overall fuel distribution and mapping process, offering potential guidance for more
effective wildfire management strategies. This section outlines the FBP fuel types, their
spatial distribution in the valley, and a correlation analysis of canopy, ladder, and understory
fuels across these types.

51

The distribution of these fuel types across the valley offers insights into the
landscape’s fire risk profile (Figure 20). While FBP System fuel types provide a broad
categorization, their low resolution and limited reliability in detailed wildfire risk
assessments have been well-documented (Baron et al., 2024). These limitations underscore
the need for more refined, high-resolution fuel mapping approaches. Despite this, the FBP
SYSTEM fuel types serve as a valuable baseline for comparison with newly developed,
higher-resolution fuel layers (Masinda et al., 2021).

52

Figure 20. Map of fuel types in the Xáxli’p Community Forest (Fountain Valley). Based on the
Canadian Forest Fire Danger Rating System (FBP System) Fire Behavior Prediction (FBP) fuel types
BC layer from 2019, provided by British Columbia Wildfire Service (BCWS) Geomatics through a
data-sharing agreement.

The relative distribution of fuel types within the valley (Figure 21) highlights the
dominance of C-7 (Ponderosa Pine/Douglas-fir) and C-3 (Mature Jack or Lodgepole Pine)
fuel types (Figure X). C-7 covers approximately 40% of the total area, primarily in the
western part of the valley, while C-3 accounts for about 30% and is concentrated in the
eastern regions. Minor fuel types, including D-1/2 (Leafless/Leafed Aspen), M-1/2
53

(Mixedwood), and C-5 (Red and White Pine), are sparsely distributed throughout the valley
and occupy much smaller areas. The O-1a/b (Grass) fuel type is more common but is
primarily being classified in alpine areas and steep slopes, often mixed with rock and scree,
as well as in some open farmland at the northern end of the study area.

Figure 21. Relative area of each fuel type, based on the Canadian Forest Fire Danger Rating System
(FBP SYSTEM) Fire Behaviour Prediction (FBP) fuel types BC layer from 2019. Provided by British
Columbia Wildfire Service (BCWS) Geomatics through a data-sharing agreement.

Canopy Fuel and Ladder Fuel. Figure 22 provides a comprehensive visualization of
the correlation between crown and ladder fuels across the FBP SYSTEM fuel types. Overall,
the correlation between these two fuel types is strong. Notably, fuel types such as O-1a/b
(Grass), C-7 (Ponderosa Pine/Douglas-Fir), and M-1/2 (Mixedwood) exhibit the highest
correlation coefficients (0.95, 0.90, and 0.90, respectively), indicating a very strong positive
relationship between crown and ladder fuels. Similarly, fuel types D-1/2 (Leafless/Leafed
Aspen) and C-3 (Mature Jack or Lodgepole Pine) show strong correlations (0.90 and 0.81,
respectively). Strong correlations in C-2 (Boreal Spruce) (0.81) are also observed.
54

Scatter plots with regression lines and box plots were generated to visualize these
relationships. The scatter plot highlights outliers and variability in ladder fuel values,
particularly in the lower range. Box plots illustrate the distribution of crown and ladder fuels
across fuel types, highlighting variances and outliers that may require special attention. The
weak correlation in C-5 (Red and White Pine) fuel type (0.48) is likely more representative
of its lack of presence in the study area rather than any real relationship with other fuels. The
overall correlation across all forest types was calculated as 0.91, further supporting the strong
relationship between crown and ladder fuels.

55

Figure 22. Correlation between canopy fuel and ladder fuel across 1000 random samples per FBP Fuel Type.

56

Canopy Fuel and Understory Fuel. There are noticeably lower correlations between
canopy and understory fuels across the FBP System fuel types, as illustrated in Figure 23.
The overall correlation coefficient between these two fuel components is -0.378, indicating a
weak negative relationship across the study area. The scatter plot highlights this trend, with a
clear negative slope in the regression line and a spread of data points reflecting variability
within and between fuel types.
Fuel-specific correlation coefficients are also presented in Figure 23, with the O-1a/b
(Grass) fuel type showing the highest correlation (0.18), though this represents only a very
weak positive relationship, likely due to the fact this fuel type has few or no trees present in
it. Conversely, most fuel types exhibit negative correlations, with D-1/2 (Leafless/Leafed
Aspen) at -0.27, C-2 (Boreal Spruce) at -0.64, C-7 (Ponderosa Pine/Douglas-Fir) at -0.55,
and C-3 (Mature Jack or Lodgepole Pine) at -0.61. Notably, the M-1/2 (Mixedwood) and C-5
(Red and White Pine) fuel types show the lowest correlations, at -0.48 and -0.42,
respectively.
The box plots and bar charts in Figure 23 provide additional insights into the
distribution of canopy and understory fuel densities across fuel types. The box plots illustrate
the variability in canopy fuel within each FBP System fuel type, with notable outliers and
differences in median values. The bar charts at the bottom of the figure emphasize the
variability in understory fuel density, further highlighting the fragmented and inconsistent
nature of this fuel layer. These visualizations reinforce the heterogeneity of understory fuels
and their weaker association with crown fuels compared to ladder fuels.

57

Figure 23. Correlation between canopy fuel and understory fuel across 1000 random samples per FBP fuel type.

58

Ladder Fuel and Understory Fuel. The scatter plot and box plots for ladder and
understory fuels show varying but generally lower degrees of correlation across the FBP
SYSTEM fuel types (Figure 24). The highest correlation is observed in the C-2 (Boreal
Spruce) fuel type with a value of 0.55, followed by the C-3 (Mature Jack or Lodgepole Pine)
fuel type with a correlation of 0.52. The C-7 (Ponderosa Pine/Douglas-Fir) and M-1/2
(Mixedwood) fuel types exhibit negative correlations of -0.41 and -0.35, respectively. The D1/2 (Leafless/Leafed Aspen) and C-5 (Red and White Pine) fuel types show the weakest
correlations, with values of -0.02 and 0.23, respectively.

59

Figure 24. Correlation between ladder fuel and understory fuel across 1000 random samples per FBP Fuel Type.

60

Aggregate Fuels and Difference Maps
The aggregated fuel maps illustrate the spatial distribution of combined fuel attributes
across the study area, highlighting regions where multiple fuel layers contribute to elevated
wildfire potential. These maps identify hotspots where high accumulations of ground, ladder,
and crown fuels occur, providing a comprehensive representation of fuel continuity across
the landscape.
Figures 25 and 26 display two aggregated fuel layers, each constructed using different
sets of normalized fuel variables. Figure 25 presents the unweighted and weighted
Aggregated Fuel Layer 1, which combines normalized values for Canopy Bulk Density
(CBD), Median Canopy Base Height (CBH), Ladder Fuel, and Understory Fuel. Figure 26
extends this aggregation by incorporating stems per hectare, adding another dimension to the
fuel assessment.
Both the unweighted and weighted maps exhibit similar spatial patterns, with high
fuel loads concentrated in specific areas. The weighting approach results in subtle variations,
as attributes with higher predictive accuracy contribute more strongly to the final aggregated
output. Regions with high aggregated scores correspond to areas with substantial
accumulations of fuel loads across different forest strata.

61

Figure 25. Normalized unweighted and weighted aggregated fuel layer 1. This map displays the spatial distribution of the unweighted aggregated fuel layer 1,
combining normalized values for Canopy Bulk Density (CBD), Median Canopy Base Height (CBH), Ladder Fuel, and Understory Fuel.

62

Figure 26. Normalized unweighted and weighted aggregated fuel layer 2. This map displays the spatial distribution of the unweighted aggregated fuel layer 2,
combining normalized values for Canopy Bulk Density (CBD), Median Canopy Base Height (CBH), Ladder Fuel, Understory Fuel, and Stems per Hectare.

63

Difference Maps
The difference maps illustrate spatial variations in fuel density across the three
primary fuel strata: canopy, ladder, and understory fuels. These maps quantify the relative
dominance of one fuel layer over another by subtracting the second fuel layer from the first.
Positive values indicate areas where the first layer has greater fuel density, while negative
values highlight regions where the second layer is more dominant. The canopy vs. ladder fuel
difference map (Figure 27) shows where canopy fuel exceeds ladder fuel (yellow to red) and
where ladder fuel is more abundant (black to dark blue). Light blue to green areas indicate
locations where both layers have similar fuel densities, suggesting a more even fuel structure.
The canopy vs. understory fuel difference map (Figure 28) highlights regions where canopy
fuel is greater than understory fuel (yellow to red) and where understory fuel dominates
(black to dark blue), indicating differences in surface and overstory vegetation density. Light
blue to green areas suggest a balanced distribution of canopy and understory fuels. The
ladder vs. understory fuel difference map (Figure 29) identifies areas where ladder fuel is
more prevalent than understory fuel (yellow to red) and vice versa (black to dark blue). Light
blue to green areas indicate relatively uniform densities of ladder and understory fuel.

64

Figure 27. Difference map between canopy and ladder fuel layers.
This map illustrates the difference between canopy and ladder fuel densities, calculated by subtracting
the ladder fuel layer from the canopy fuel layer. Positive values indicate areas where canopy fuel
density is greater than ladder fuel density, whereas negative values highlight areas where ladder fuel
is more abundant than canopy fuel. The color scale ranges from black and dark blue, representing
regions with higher ladder fuel dominance, to red and orange, indicating areas with a more canopydominant structure.

65

Figure 28. Difference map between canopy and understory fuel layers.
This map presents the difference between canopy and understory fuel densities, with values
calculated by subtracting the understory fuel layer from the canopy fuel layer. Positive values
represent regions where canopy fuel density exceeds understory fuel density, whereas negative values
indicate areas where understory fuel is more dominant. The visualization provides insights into
locations where dense canopy cover coincides with sparse understory vegetation and vice versa,
affecting fire spread dynamics.

66

Figure 29. Difference map between ladder and understory fuel layers.
This map displays the spatial differences between ladder and understory fuel densities, generated by
subtracting the understory fuel layer from the ladder fuel layer. Positive values denote areas where
ladder fuel is more dominant than understory fuel, suggesting increased potential for fire transition
between ground and canopy fuels. Negative values highlight regions where understory fuels exceed
ladder fuels, indicating areas where fire behavior may be more concentrated at ground level rather
than propagating vertically.

67

Discussion
Key Fuel Maps
The spatial distribution of canopy, ladder, and understory fuels provides valuable
insights into forest fuel stratification and fire behavior potential (Viedma et al., 2024).
Canopy fuel, which is densest in mature forest stands, suggests areas with high crown fire
potential. Ladder fuels, which are highly variable across the study area, highlight locations
where surface fires may transition to the canopy, increasing the likelihood of high-intensity
crown fires. Similarly, understory fuels contribute to surface fire spread, with their density
influencing the potential for ignition and flame height (Forbes et al., 2022).
The presence of high-density ladder fuels beneath areas of continuous canopy fuel
may indicate regions more susceptible to crown fire initiation and spread. Conversely, low
ladder fuel density beneath areas of high canopy fuel may suggest that surface fires in these
areas are less likely to transition to crown fires. These insights underscore the importance of
assessing vertical fuel continuity when evaluating fire behavior risk (González‐Ferreiro et al.,
2017; Schoennagel et al., 2012).
Considering these fuel layers along with supplementary attributes such as stems per
hectare, median canopy base height, and canopy bulk density (a common metric very similar
to the custom canopy fuel metric in this study), fire risk assessment becomes more
comprehensive. A key finding is that these additional attributes can be used to build strong
predictive models, which has broader implications for other forest-related activities. Stands
with high stem density and low median canopy base height may experience more intense fire
behavior due to increased fuel availability and fire laddering potential. Similarly, areas with

68

high canopy bulk density and low ladder fuel density may sustain crown fires but may be less
likely to facilitate surface-to-crown fire transitions.
These findings support further analysis using aggregated fuel layers and difference
maps to better understand how these fuel layers interact and influence fire behavior, allowing
for a more multi-faceted assessment of wildfire susceptibility across the landscape. This
involves examining spatial variability, understanding fuel interactions (e.g., how ladder fuels
facilitate fire spread), and using difference maps to identify variations in fuel densities across
different fuel layers, helping to refine fire risk assessments.
Random Forest Model and ALS Performance
The Random Forest model's stronger performance in predicting canopy fuel
attributes, compared to understory and ladder fuels, highlights the capabilities and limitations
of the approach. One primary limitation lies in the ability of Airborne Laser Scanning (ALS)
sensors to penetrate dense forest canopies. ALS beams often reflect off the upper components
of the forest structure, resulting in fewer data points for lower layers like the understory and
ladder fuels. This issue is further compounded by the less precise and lower quantity of field
data available to train the models for these strata, especially where data collection methods
relied on density and cover percentage estimations rather than direct measurements.
Despite these challenges, the strong performance for canopy-related attributes
demonstrates ALS's effectiveness in capturing structural characteristics of the canopy fuel
layer of the forest fuel strata (Reilly et al., 2021). This is likely due to the more consistent
and detectable nature of canopy features compared to the heterogeneity and limited visibility
of understory and ladder fuels. Improving the accuracy of models for these lower forest
layers may require integrating additional data sources or refinements to model parameters.
69

While terrestrial ALS could provide more detailed insights into understory and ladder
fuels, its limited feasibility for covering large areas, such as the 34,000 hectares of the
Xáxli’p Survival Territory, renders it impractical for this study's scope (Hopkinson et al.,
2004; Liang et al., 2018). Nevertheless, despite reduced precision, ALS-derived layers for
understory and ladder fuels remain valuable for operational and community-focused
applications particularly when acceptable models for fuel densities are developed. Recent
studies suggest that combining ALS with optical imaging sensors enhances fuel mapping
capabilities across all forest layers (Bright et al., 2017; Chen et al., 2016; Wulder et al.,
2016). This integrated approach leverages the strengths of both technologies, offering a
promising avenue for improving the accuracy and reliability of fuel maps, particularly in
challenging strata like understory and ladder fuels.
Correlation Between ALS generated Fuels and FBP Fuel Types
When considering the relative area covered by each of the FBP System fuel types in
the Xáxli’p Survival Territory, it is useful to note the geographic segregation of fuel types
(Figure 20). C-7 predominantly occupies the western part of the valley, and C-3 is more
concentrated in the eastern regions, which suggests the need for targeted fire management
strategies that consider the specific characteristics of these fuel types. However, given the
limited reliability of the FBP System layer, these results should be interpreted with caution
(Baron et al., 2024). Considering the lack of local scale resolution and reliability of the FBP
SYSTEM layer noted in the Xáxli’p Survival Territory, relying on the FBP System fuel maps
to develop management strategies could exacerbate fire risk and worsen a bad situation,
however its important to acknowledge that where no other information about fuels exists
some information is certainly better than no information.
70

It's important to note that although we are inferring fuel density from the ALS data we
cannot obtain vegetation composition from the ALS so comparing to fuel types which
consider vegetation composition has inherent challenges. That being said figure 21 indicates
that D 1/2, M 1/2, and C-5 fuel types are present in minor quantities throughout the valley.
Due to their limited presence, the classification of these fuel types is likely unreliable, and
their existence does not necessarily indicate any significant difference in the fuels in these
areas. Additionally, while the O-1a/b fuel type is more abundant, it is also prone to error. This
classification also captures alpine areas that include large rock and scree components, as well
as rock and scree slopes on steeper sections of the valley. Although it correctly classifies
some grassy fields in the valley bottoms through observation, investigation, and familiarity
with the territory, it is notably prone to error in many locations and should not be considered
reliable. Therefore, the following correlation analysis will primarily focus on the more
abundant fuel types in the valley, specifically C-7, C-3, and C-2.
The correlation analysis shows a strong relationship between canopy and ladder fuel
densities in the dominant fuel types of the valley: C-7 (Ponderosa Pine/Douglas-Fir), C-3
(Mature Jack or Lodgepole Pine), and C-2 (Boreal Spruce). Higher canopy fuel density
generally corresponds with higher ladder fuel density, increasing the potential for vertical fire
spread and overall fire hazard (Reilly et al., 2021).
C-2, C-3, and C-7 all exhibit relatively high fuel densities in both layers, with
significant variation across observations. However, C-7 shows slightly greater variability,
particularly in canopy fuel density, likely due to differences in stand age, species
composition, and disturbance history. This variability suggests that while fuel loads in C-7
are generally high, local factors may influence fire behavior more than in C-2 or C-3.

71

These findings emphasize the need for targeted fuel management to reduce the high
horizontal and vertical densities in some areas of the study area. Reducing ladder fuels can
help mitigate crown fire risk in all three fuel types; however, the higher variability in C-7
suggests that treatment strategies should be more adaptive to local conditions, focusing on
ways to pin point these fuel types of key concern and reduce hazard through thinning and
targeted fuel management. The strong correlation between canopy and ladder fuel densities
across these fuel types highlights the importance of understanding vertical fuel structures
when assessing fire behavior potential.
Several ecological factors may explain the lower correlations between canopy and
understory fuel densities. Increased light availability can promote understory vegetation
growth in areas with lower canopy fuel loads, potentially increasing understory fuel density
and creating a negative correlation between the two layers. Unlike canopy fuels, understory
fuels are often more fragmented and inconsistent, with disturbances such as a treefall forming
dense fuel patches that contribute to overall heterogeneity (Ma et al., 2024).
Environmental features like rocks and boulders can further impact fuel density
estimates by interfering with airborne laser scanning (ALS) measurements. As previously
discussed, ALS has limitations in densely forested areas, where laser pulses may not
penetrate deeply enough to capture understory fuels accurately. These measurement
challenges and the natural variability of understory vegetation, shaped by microhabitat
conditions, species composition, and disturbances, add to the complexity of assessing fuel
distributions (Maltamo et al., 2020).
By highlighting these inconsistencies, this analysis not only refines our understanding
of local fuel density patterns but also helps assess the effectiveness of the FBP System in

72

capturing fuel densities at finer spatial scales. This is critical for improving local wildfire risk
mapping and informing more precise fuel treatment planning, ensuring that mitigation
strategies are based on the true variability of fuel loads in different forest conditions (Hanes
et al., 2021).
Several factors may explain the lower correlations between ladder and understory fuel
densities. Ladder fuels, comprising smaller trees and shrubs from approximately chest height
(1.37 meters DBH) to the median canopy base height,do not always correspond to dense
understory vegetation, which includes vegetation from ground level to chest height. In areas
with dense canopy and ladder fuel layers, understory growth may be suppressed due to
reduced light availability, further contributing to the variability in understory fuel density.
Conversely, in more open areas with lower canopy and ladder fuel densities, increased light
penetration can promote understory vegetation growth, potentially leading to higher
understory fuel loads. The variability in understory fuels is also influenced by microhabitat
conditions, species composition, and disturbances such as treefall, which create localized fuel
accumulation and add to the heterogeneity of this fuel layer (Hiers et al., 2009). Additionally,
ALS limitations in densely vegetated areas may reduce the accuracy of understory fuel
estimations, as laser pulses struggle to penetrate deep enough to capture finer-scale fuel
distributions (Hilker et al., 2010).
Despite these limitations, it’s also important to acknowledge that overall the results of
this study are mainly positive and the model performance was relatively good. ALS-derived
fuel mapping still offers significantly higher-resolution assessments of fuel variability
compared to the broad fuel type classifications of the FBP SYSTEM, especially when all
ALS-derived fuel layers, canopy, ladder, and understory, are considered together (Crotteau et

73

al., 2019). However, the reliability of ALS-based mapping is highest for canopy fuels,
whereas the accuracy of ladder and understory fuels is more variable due to penetration
challenges. In areas with dense canopy and ladder fuel layers, it is ecologically likely that
understory fuel is also high. Still, ALS cannot fully capture these lower strata due to
obstruction by upper layers. Conversely, in areas where canopy density is lower, ALS is more
likely to effectively penetrate and provide a more reliable assessment of ladder and
understory fuel densities (Hilker et al., 2010).
These challenges must be carefully considered when using ALS-derived fuel layers to
inform fuel treatment planning, risk identification, and mitigation efforts. While ALS data
improves spatial resolution compared to traditional classification systems, understanding its
limitations, particularly for lower fuel strata, ensures that fire behavior models and
management strategies are built on a more accurate representation of fuel distribution.
Integrating ALS-derived canopy fuel data with ladder and understory layers can provide a
more holistic understanding of vertical fuel continuity. However, caution should be exercised
in heavily forested areas where lower strata are more difficult to map reliably.
Aggregate Fuels and Difference Maps
Aggregate Maps
The aggregation of fuel layers provides a higher resolution view of wildfire hazard by
combining multiple fuel attributes into a single index. This approach allows for the
identification of areas where high concentrations of multiple fuel types coincide, which is
critical for understanding the cumulative effects of fuel accumulation on fire risk (Cameron
et al., 2021; Forbes et al., 2022; Mallinis et al., 2016). The unweighted aggregates provide a
broad measure of overall fuel load, treating each fuel layer equally, whereas the weighted
74

aggregates incorporate predictor reliability, ensuring that attributes with greater predictive
strength contribute more significantly (Figures 25 and 26).
By incorporating weighting based on cross-validation R² values, the aggregated layers
integrate a statistical measure of model performance, allowing the final outputs to reflect not
only fuel load estimations but also confidence in the underlying predictions. The unweighted
aggregation method assumes that all predicted fuel layers contribute equally to overall fire
risk, regardless of how well the Random Forest models performed in predicting those layers.
In contrast, the weighted aggregation method adjusts for variability in model accuracy by
scaling each predicted fuel layer according to its R² value, ensuring that layers with stronger
predictive reliability have greater influence on the final aggregated fuel layer. This approach
enhances the statistical robustness of the final output, allowing decision-makers to consider
the relative confidence in different predicted fuel layers when assessing overall wildfire risk,
gaining a more nuanced and comprehensive perspective on the reliability and validity of
these layers.
It is important to acknowledge, however, that of the three fuel layers considered,
canopy, ladder, and understory, surface fuels were not included, despite their potentially
critical role in driving fire behavior (Kelly et al., 2017). This exclusion was not due to a lack
of importance but rather the difficulty in accurately modeling surface fuels with the available
data. As a result, this study relies on the canopy, ladder, and understory fuel layers as proxies
for overall wildfire risk, recognizing that surface fuels may, in reality, be the most influential
but remain challenging to incorporate with current remote sensing and modeling capabilities.
The concept of validation is further expanded in Chapter Three; however, in this
context, integrating predictive strength into the aggregate layers enables a more informed

75

assessment of confidence levels associated with these layers. On this note, it is important to
acknowledge that these are modeled and predicted resultant layers rather than direct
representations of reality. However, they serve as highly effective estimation mapping
products, offering high-resolution fuel density insights across the entire landscape by
leveraging ALS’s advanced remote sensing capabilities combined with field data collection
techniques and the Random Forest machine learning modeling approach (Pierce et al., 2012;
Xu et al., 2024).
The spatial patterns in the aggregated fuel maps highlight regions where canopy,
ladder, and understory fuels align, forming continuous vertical fuel structures that increase
the inherent wildfire hazard (Parsons et al., 2017; Benali et al., 2021). These higher-risk areas
should potentially be considered higher priority for fuel management interventions, such as
thinning, prescribed burns, and fuel breaks, to reduce vertical connectivity and mitigate fire
risk. Conversely, areas with low aggregated fuel density values may represent lower-risk
regions where fire spread potential is reduced due to discontinuous fuels. However, these
areas should still be monitored, as changes in stand structure over time may alter their fire
susceptibility.
Furthermore, fuel treatment and wildfire risk mitigation decisions should not be based
solely on fuel density or concentration. Additional factors must be considered, including
proximity to values at risk, dominant weather patterns, local topography, fuel connectivity,
and directional wildfire vulnerability assessments (Beverly & Forbes, 2023). Fuel treatments
must also be balanced against practical constraints, as these efforts are inherently time and
resource-intensive, particularly for smaller communities with limited capacity. In some cases,
focusing on areas with lower fuel density may be more feasible and impactful, as it allows

76

for larger-scale treatments that can still reduce fire risk meaningfully. A multi-faceted
approach to prioritization is essential to ensure that risk reduction efforts are strategically
allocated where they will have the greatest overall benefit (Domingo et al., 2020; Yao et al.,
2020).
Despite these complexities, the value of high-resolution fuel data in supporting
informed management decisions is undeniable. More detailed spatial fuel information
enables a nuanced understanding of fuel distribution, helping to refine risk assessments and
optimize mitigation strategies. Additionally, it is also crucial to consider the temporal
resolution of fuel data, as forests naturally grow and change over time. The fuel density
estimates in this study represent a snapshot of the current forest structure, serving as a
valuable reference point for future monitoring and adaptive management. Ensuring that fuel
data is updated and integrated into long-term planning efforts will be essential for
maintaining effective wildfire risk mitigation strategies.
Beyond structural characteristics, fuel composition also plays a critical role in fire
behavior. Multi-spectral data can be leveraged to enhance fuel classification, particularly in
distinguishing between coniferous and broadleaf species, which have significantly different
flammability characteristics. Since conifers generally burn more intensely and contribute to
crown fire spread, integrating spectral information with ALS-derived fuel metrics could
further improve fire risk assessments by providing insights into vegetation type alongside
fuel density (Ager et al., 2010). Expanding future analyses to include multi-spectral data
would offer a more complete picture of fuel conditions and improve the accuracy of wildfire
susceptibility models.

77

Difference Maps
The difference maps provide valuable insight into the spatial variability of fuel
stratification, emphasizing the importance of understanding vertical fuel continuity beyond
individual layers. By comparing the relative dominance of canopy, ladder, and understory
fuels, these maps help identify areas where fire is more likely to transition between fuel
strata, a key factor in predicting wildfire spread and intensity. Supplementing this
information with forest and vegetation type data would further enhance the ability to predict
wildfire hazard, as different vegetation types have distinct fuel characteristics that influence
fire behavior. Integrating this additional layer of information could improve fire modeling,
helping to refine risk assessments and guide more effective mitigation strategies. The results
reinforce the value of using multiple vertical fuel strata layers and integrating this nuanced
perspective of fuel layers when assessing fire risk, rather than relying on a single variable, as
vertical fuel connectivity significantly influences fire behavior (Vafaei et al., 2018; White et
al., 2016).
The canopy minus ladder fuel difference map (see Figure 27) reveals areas where
canopy fuel is significantly greater than ladder fuel, suggesting locations where fire may have
difficulty transitioning from surface to crown fires unless driven by strong winds or embers.
This observation underscores the critical role of fuel connectivity in fire behavior (Allen et
al., 2023). These areas may be less susceptible to rapid vertical fire spread but could still
sustain crown fires under extreme weather conditions. Conversely, areas where ladder fuel is
more prominent than canopy fuel indicate locations where fire can more easily climb from
the ground to the upper canopy, increasing the likelihood of crown fires.

78

The canopy minus understory fuel difference map highlights regions where the upper
canopy is dense, but understory fuel is sparse. In these areas, ground fire intensity may be
lower due to reduced fine fuel availability, but crown fire potential remains, particularly
under strong wind conditions. Conversely, areas where understory fuel exceeds canopy fuel
indicate locations with high surface fuel loading, where intense surface fires could occur but
may have a lower probability of transitioning to crown fires unless ladder fuels provide
sufficient connectivity (Forbes et al., 2022; Raymond & Peterson, 2005). These patterns
suggest that fuel-treatment strategies should be tailored to address both surface and canopy
fuel connectivity, particularly in regions where fuel discontinuities create complex fire
behavior dynamics.
The ladder minus understory fuel difference map identifies regions where ladder fuels
dominate relative to understory fuels, forming a continuous vertical fuel pathway that
facilitates the transition of surface fires into the canopy. These areas are particularly critical
for understanding fire behavior and risk assessment, as they indicate where ladder fuels act as
a bridge between the understory and upper canopy. Conversely, areas where understory fuels
are dominant, but ladder fuels are sparse suggest a lower likelihood of vertical fire spread,
though they may still support intense understory fire behavior. These results reinforce the
value of using multiple vertical fuel strata layers and integrating this nuanced perspective of
fuel layers when assessing fire risk, rather than relying on a single variable, as vertical fuel
connectivity significantly influences fire behavior.
As previously mentioned, it is important to reiterate that the ability or inability of
ALS to penetrate through dense upper canopy layers influences the accuracy of lower fuel
strata measurements. This difference mapping approach provides a useful tool for identifying

79

areas where ALS-derived estimates of lower fuel layers may be more or less reliable due to
the density of overlying fuel strata. For example, in the canopy minus ladder fuel map, areas
with lower values, where canopy fuel is less than ladder fuel, may indicate locations where
ALS can more effectively penetrate to capture ladder fuel, increasing the reliability of ladder
fuel estimates. Conversely, areas with higher values, where canopy fuel dominates over
ladder fuel, suggest that ALS penetration was likely reduced, meaning ladder fuel
measurements in these regions may be less reliable.
Similarly, in the canopy minus understory fuel map, lower values may indicate
locations where the canopy is more open, allowing for better ALS penetration and more
reliable understory fuel estimates. In contrast, higher values suggest dense upper canopy
cover, which likely obstructed ALS pulses from reaching the forest floor, reducing the
accuracy of understory fuel measurements. The ladder minus understory fuel map follows the
same principle, where ladder fuel is greater than understory fuel, ALS penetration to the
ground layer may have been limited, making understory fuel estimates less reliable.
Conversely, where understory fuel is dominant, ALS may have had better visibility of these
fuels, improving measurement accuracy.
These insights highlight the importance of considering ALS penetration limitations
when interpreting fuel difference maps. While these maps provide valuable spatial context
for understanding vertical fuel continuity, they should also be used to assess the confidence
associated with lower fuel layer estimates based on the density of overlying fuel strata.
Integrating this awareness into fire risk assessments and fuel management planning can help
refine decision-making by recognizing areas where fuel data may be more or less
representative of actual conditions.

80

Traditional risk classifications often generalize fuel conditions over large areas.
However, the aggregated and difference maps in this study highlight finer-scale variations in
fuel loading and connectivity, improving the accuracy of fire behavior models and enhancing
decision-making for fuel treatment planning (White et al., 2016; Zhu et al., 2024).
By quantifying these differences, the results highlight the potential value in designing
localized and adaptive fuel management strategies that address specific fuel configurations
rather than relying on broad, landscape-scale treatments. Additionally, in interpreting these
results, it is important to consider factors beyond fuel density alone when planning wildfire
mitigation strategies. While high-resolution ALS-derived fuel data provides detailed spatial
insights, effective risk reduction efforts must integrate additional considerations such as
proximity to values at risk, dominant weather conditions, local topography, fuel connectivity,
and directional wildfire vulnerability assessments (Beverly & Forbes, 2023; Spits et al.,
2017; Wang & Niu, 2016). These broader landscape and environmental factors play a critical
role in shaping fire behavior and must be incorporated alongside fuel density data to develop
comprehensive and effective mitigation strategies.
Finally, while these maps offer significant advantages for wildfire risk assessment, it
is crucial to recognize that they represent a snapshot in time of current fuel structures. Forests
naturally evolve due to growth, disturbances, and succession, meaning fuel distributions
change over time (Stephens et al., 2009). As such, ongoing monitoring and updates to fuel
data will be essential for maintaining accurate fire risk assessments and informing long-term
mitigation strategies. The ability to incorporate updated high-resolution fuel data into
management decisions will help ensure that wildfire risk reduction efforts remain effective
and adaptive in response to changing forest conditions.

81

Conclusion
Identifying high-risk fuel hotspots has important implications for wildfire
management. By recognizing areas where multiple fuel layers elevate fire risk, fire managers
can prioritize resource allocation and focus fire risk mitigation and prevention efforts more
effectively. This helps to promote fuel reduction treatments, firebreak construction, and
community preparedness programs that are focused on the most critical and impactful
regions. Targeting high-risk hotspots allows for more efficient use of limited resources,
yielding significant benefits in reducing fire incidence and mitigating potential damage.
Additionally, understanding the spatial convergence of hazardous fuel layers helps in
developing comprehensive mitigation strategies that address multiple risk factors
simultaneously.
This study demonstrated that ALS-derived fuel data, combined with Random Forest
modeling, provides a powerful and detailed means of characterizing vertical and horizontal
fuel distributions. Canopy fuels were mapped with high accuracy, reflecting ALS’s strength
in capturing overstory structure. On the other hand, ladder and understory fuels showed
greater variability due to ALS penetration limitations and the inherent heterogeneity of these
lower fuel strata. Despite these challenges, integrating weighted aggregation layers improved
interpretability by incorporating predictor reliability, allowing for a more informed
assessment of high-risk areas. The spatial patterns of canopy, ladder, and understory fuels
revealed important structural differences across the landscape, reinforcing the importance of
evaluating vertical fuel continuity rather than relying on a single fuel variable.
This study highlighted significant within-class variation in fuel loading by comparing
ALS-derived fuel layers with the FBP System fuel types. While the FBP System provides a

82

useful broad-scale classification system, and provides fire-behavior estimates calculated
using weather and topography (Baron et al., 2024), it does not adequately account for
localized fuel variability (Groot et al., 2022). The results showed that C-7, C-3, and C-2 fuel
types, which dominate the study area, contain substantial internal differences in canopy,
ladder, and understory fuel densities, suggesting that relying solely on FBP System
classifications could overlook important fire risk factors at finer spatial scales. Correlation
analyses further revealed that canopy and ladder fuels exhibited strong relationships in
dominant fuel types, underscoring the role of ladder fuels in facilitating crown fire
transitions. In contrast, the weaker correlation between canopy and understory fuels, as well
as ladder and understory fuels, reflected the complex ecological factors influencing fuel
stratification, including light availability, species composition, and disturbance history
(Kreye et al., 2014).
The application of ALS-derived difference maps provided a novel approach to
assessing vertical fuel connectivity, identifying areas where transitions between fuel strata are
most likely. These maps not only revealed regions with strong ladder-to-canopy connectivity,
where crown fire potential is greatest, but also areas where ALS penetration limitations may
affect the accuracy of lower-strata fuel estimates. The development of weighted and
unweighted aggregate fuel layers further refined wildfire risk assessments by integrating
multiple fuel attributes into a single index. The weighted approach, which accounts for model
confidence, provided a more nuanced representation of fuel distribution, helping forest
managers assess fuel loading and prediction reliability simultaneously. These improvements
demonstrate how ALS and machine learning can enhance fire risk classification, offering
data-driven decision support for wildfire mitigation planning.

83

The detailed forest fuel maps and analytical insights generated in this study have
direct applications for wildfire risk management in the Xáxli’p Survival Territory. By
identifying areas with high densities of all three vertical fuel strata, forest managers can
prioritize fuel treatments such as thinning, pruning, and prescribed burns to disrupt vertical
and horizontal fuel continuity. Also, identifying high-risk zones through spatial clustering
analysis ensures that risk reduction efforts can be strategically allocated to areas where
interventions will have the greatest impact.
However, as emphasized throughout this study, fuel management must extend beyond
fuel density data alone. Effective mitigation requires a multi-faceted approach that
incorporates additional factors such as proximity to values at risk, dominant weather patterns,
local topography, fuel connectivity, and directional wildfire vulnerability assessments
(Beverly & Forbes, 2023; Paveglio et al., 2016). Moreover, practical constraints, including
resource limitations and operational feasibility, mean that fire risk management decisions
must balance scientific precision with real-world logistics, particularly for smaller
communities with limited fire management capacity.
Another key consideration is the temporal nature of fuel distributions. Forests
continuously evolve due to growth, disturbances, and succession, meaning that any static fuel
map represents only a snapshot in time. Future research should explore methods for updating
ALS-derived fuel data through multi-temporal ALS acquisitions, integrating forest growth
modelling approaches, remote sensing fusion techniques, and ongoing field validation.
Incorporating data on ground fuels, which were not measured in this study but play a critical
role in fire dynamics, could further enhance wildfire risk models.

84

Combining ALS remote sensing technologies, machine learning, and local knowledge
provides a more comprehensive and operationally relevant approach to wildfire risk
assessment. These findings highlight the strengths and limitations of ALS-derived fuel data,
demonstrating its value for improving fire risk classification, refining spatial fuel
assessments, and guiding targeted mitigation efforts. These results offer practical,
community-driven solutions for fire management in the Xáxli’p Survival Territory and
similar landscapes. The insights gained in this chapter provide a foundation for Chapter 3,
which explores structured empirical measurements and opportunistic ocular estimates,
building upon the integration of advanced mapping techniques and field-based validation
discussed here. This transition reinforces the importance of combining quantitative data with
qualitative insights to improve the robustness and applicability of wildfire risk management
strategies, particularly by incorporating local context and community-based knowledge to
ensure that risk assessments and mitigation efforts are ecologically, culturally, and locally
relevant.

85

Chapter 3: Comparing Sampling Techniques in Forest Fuel Mapping: Structured
Empirical vs. Opportunistic Ocular Estimates with ALS and Random Forest
Introduction
As outlined in Chapter 2, accurate forest fuel mapping is essential for managing
wildfire risks and supporting strategic planning and real-time decision-making in forest
management. The increasing frequency and severity of wildfires globally, driven by a
changing climate and anthropogenic factors, underscore the importance of accurate forest
fuel mapping (Flannigan et al., 2009; Jolly et al., 2015). Traditionally, forest planners and
managers assessed fuel loading and implemented reduction strategies based on broad
observations of forest types and general fire risks (Agee & Skinner, 2005). These evaluations
were often implicit and subjective, varying with perceived threat levels, and they relied
heavily on ground-based observations and manual data collection. While valuable, these
approaches were limited in scope and scalability (Keane et al., 2001).
With increasing wildfire risks, there has been a shift toward explicitly integrating fire
risk assessment, which requires better data on fuel loading (Laushman et al., 2020). Ideally,
this data is quantitative, collected at an appropriate spatial resolution, and allows for distinct
evaluation of different fuel types and structural categories. The advent of Airborne Laser
Scanning (ALS) and other remote sensing technologies has revolutionized forest fuel
mapping by enabling detailed and extensive data collection over large areas with high
precision, thus reducing the reliance on intensive field sampling (White et al., 2016; Wulder
et al., 2012).
Several frameworks can be used to generate fuel load data and apply it to a
landscape-level assessment of wildfire fuel load and wildfire risk. These frameworks can be
86

broadly categorized into three main approaches: field-based methods, remote-sensing
techniques, and hybrid integration techniques. Field-based methods rely on ground
measurements to capture detailed local and site-specific data and measurements relating to
forest structural characteristics. Remote sensing techniques, such as ALS and satellite
imagery, enable extensive data collection over larger areas with increasing precision. Hybrid
approaches combine these methods, integrating ground and remote sensing data with
advanced modelling techniques to enhance accuracy, scalability, and efficiency (Forbes et al.,
2022; He et al., 2024; Skowronski et al., 2015).
This chapter compares two distinct field data collection techniques for assessing
forest fuel loading: structured empirical (SE) measurements and ocular estimates (OE). These
methods are evaluated alongside ALS and Random Forest (RF) modeling to assess their
effectiveness in producing accurate wildfire fuel maps. Both SE and OE contribute valuable
datasets for training landscape-scale models that integrate remote sensing data to predict fuel
load distributions. By combining field measurements with advanced remote sensing
technologies like ALS and predictive tools like RF modeling, these approaches enhance
understanding of wildfire fuel dynamics across diverse forest landscapes.
Structured empirical (SE) measurements and ocular estimates (OE) have distinct
strengths and limitations. Structured empirical measurements involve collecting detailed
physical data within predefined plots using a strict random stratified sampling framework. As
outlined in Chapter 2, key attributes measured include tree height, species, diameter at breast
height (DBH), and crown base height (CBH). This method relies on specialized tools such as
hypsometers and allometric calculations, requiring technical expertise, time, and resources
(Bright et al., 2017; Cruz et al., 2003; FP Innovations, 2020; Government of British

87

Columbia, 2021; Phelps & Beverly, 2022). While SE provides high precision, its intensive
nature limits the number of samples collected, reducing the area covered within a given
timeframe. For further details on SE methods, refer to Chapter 2.
In contrast, OE relies on visual estimates of vegetation density across different height
intervals. Conducted by local forest crew members with in-depth landscape knowledge, this
method enables rapid and flexible data collection without specialized equipment or extensive
training. The efficiency of ocular estimates allows for many more samples over a broader
area in less time, making it particularly advantageous for landscape-level analyses. While
OE's reliance on individual judgment introduces variability and reduces precision compared
to SE, its ability to rapidly gather extensive data provides critical insights for wildfire risk
assessment, particularly in resource-constrained or time-sensitive scenarios.
Machine learning algorithms require large, diverse training datasets with minimal
measurement errors to produce reliable landscape-level projections. Capturing the full
spectrum of forest types ensures model generalization and improves accuracy across different
forest structures and fuel conditions (Pichler et al., 2023; Stupariu et al., 2022). This
framework enables a robust comparison of SE and OE, assessing their effectiveness in
landscape-scale wildfire risk modeling.
By evaluating the potential of community-driven data collection, this study aims to
advance wildfire risk mapping that is accessible, reliable, and actionable. Community-driven
approaches build capacity, foster self-reliance, and strengthen resilience, supporting climate
adaptation at local and regional scales (Castleden et al., 2012). Engaging local communities
not only enhances data relevance but also instills ownership and empowerment, which is
particularly relevant for Indigenous communities (Copes‐Gerbitz et al., 2022).

88

To assess the potential of community-driven data collection, this study examines SE
and OE by addressing key research questions: What are the trade-offs between SE’s precision
and resource demands versus OE’s accessibility and speed? Which method produces more
robust models for essential wildfire fuel metrics, Canopy Fuel, Ladder Fuel, and Understory
Fuel? What unique strengths does each approach offer? By exploring these questions, this
study evaluates the effectiveness of SE and OE in wildfire fuel mapping and identifies
opportunities to refine sampling strategies, ultimately making fuel assessments more
efficient, scalable, and accessible, particularly for communities with limited resources.
Through this comparison, we aim to support the development of practical, data-driven
wildfire management approaches that balance accuracy, efficiency, and inclusivity.
Methods
Study Area, Data, and Fuel Layers
As outlined in Chapter 2, this study was conducted within the Survival Territory of
the Xáxli’p First Nation, encompassing the Xáxli’p Community Forest. Located in the
Pavilion Ranges eco-section within the traditional territory of the St'át'imc Nation, east of
Lillooet, British Columbia, this region, commonly referred to as the Fountain Valley, spans
approximately 35,000 hectares (Demarchi, 2011; Forest Analysis and Inventory Branch,
2022). For further details on the study site, see the subsection “Study Area” under Chapter 2.
Airborne Laser Scanning (ALS) data was collected across the Xáxli’p Survival
Territory to provide high-resolution, three-dimensional information on forest structure. The
ALS point cloud data was processed to extract metrics relevant to canopy, ladder, and
understory fuel layers. These metrics included point density values binned by height
intervals, which were used to calculate biomass distribution and fuel loads. Consistent
89

preprocessing steps ensured compatibility between SE and OE datasets, enabling reliable
comparisons and landscape-level predictions.
Three key fuel layers, Canopy Fuel (CF), Ladder Fuel (LF), and Understory Fuel
(UF), were identified by the community as critical for understanding fuel distributions and
supporting wildfire risk assessment, planning, and mitigation efforts. Understanding forest
fuels in the context of these three fuel pools is fundamental to understanding fire behavior
and fire risk modelling, as the distribution of fuel loads across these vertical strata directly
influences the probability and intensity of fire spread across the landscape (Ivey et al., 2024;
Alcasena et al., 2017). They are also impacted differently by various forest management
actions, such as thinning treatments, prescribed fire, and cultural burning, which alter fuel
composition and distribution in distinct ways (Agee & Skinner, 2005; Scott & Reinhardt,
2001). These distinct fuel layers are essential for evaluating wildfire risk and informing
strategies to reduce those risks while supporting ecological health and community objectives.
The definitions of these three key fuel layers remain consistent with those in Chapter
2. Understory fuel includes low-lying vegetation such as grasses, shrubs, and small trees,
measured from ground level to approximately 1.37 meters. It is important to note that
understory fuel is distinct from ground fuel, which consists of materials such as leaf litter,
duff, and decomposing organic matter, fuel sources that are largely inaccessible to ALS
(Bright et al., 2017). Ladder fuel is a vertical link between the understory and canopy,
consisting of smaller trees, branches, and tall shrubs that provide a pathway for surface fires
to move upward. Ladder fuel density ranges from 1.37 meters (breast height) to the median
canopy base height. Canopy fuel refers to the density of vegetation in the uppermost sections
of the forest, extending from the median canopy base height to the tops of the canopies and

90

including upper stems, branches, leaves, and needles. In the Fountain Valley, the minimal
presence of deciduous cover reduces the need to distinguish between mixed-cover fuels.
Canopy fuel plays a critical role in determining the potential for high-intensity wildfires that
spread rapidly through tree canopies and can be highly destructive (Cruz et al., 2003).
The two methods for sampling, structured empirical (SE) and opportunistic ocular
estimate (OE), are introduced in detail in this section. While these approaches differ, specific
steps were taken to ensure their outputs could be effectively compared. The SE dataset
comprised 104 samples, while the OE dataset included 695 samples, providing much greater
spatial coverage. SE data were continuous and normalized to a scale of 0 to 1, whereas OE
data were collected as discrete categories (1, 2, 3, 4, 5). To ensure consistency in the
modelling approach, the same Random Forest regression framework was applied to both
datasets, treating the OE discrete categories as a continuous scale for modelling purposes. All
fuel layer metrics were normalized to a 0–1 scale, allowing the outputs of both methods to be
directly compared despite differences in data collection techniques.
The OE data, collected using discrete categories (1–5), were treated as continuous for
the Random Forest regression modeling and statistical analysis. This decision was consistent
with assumptions made during field data collection, where the ordinal categories were
intended to approximate a continuum of fuel density. The field crew operated under the
assumption that intervals between categories (e.g., the difference between 1 and 2 was
equivalent to the difference between 4 and 5) represented equal gradients in fuel density. This
approach ensured that visual estimates aligned with the underlying variability of fuel loads
observed in the landscape.

91

We acknowledge that this assumption may introduce minor biases, particularly in
correlation testing where the discrete nature of the data might affect sensitivity. However, this
approach was deemed appropriate to maintain consistency across datasets and facilitate
unified analysis. The implications of this decision are further explored in the results and
discussion sections.
Structured Empirical (SE) Sampling
As covered in Chapter 2, the SE method employed a random stratified sampling
design to ensure representation of different fuel types and elevations across the landscape. A
total of 104 sample plots were collected by the Xáxli’p forest crew and project team over the
course of four weeks, each with an 11.28-meter radius. Figure 16 shows the distribution of
sample plots throughout the study area. To further enhance data collection, 3-meter diameter
subplots were positioned at the four cardinal directions around the circumference of the main
plot. See Figure 16 in Chapter 2 for a schematic of this sample plot design. These subplots
gathered supplemental visual estimate data, adding additional information.
At each SE plot, detailed physical measurements were taken, including tree heights,
species identification, diameter at breast height (DBH), and crown base height (CBH). Tree
heights were measured using hypsometers, which calculate the vertical distance to the top of
a tree by combining angle readings and distance measurements from the observer to the base
of the tree, ensuring precise and repeatable results. The DBH was determined using
calibrated diameter tapes to account for the circumference-to-diameter conversion, wrapped
tightly around each tree at 1.37 meters above ground level for consistency. Species
identification relied on visual inspection and botanical keys to ensure accurate classification
of individual trees. Crown base height was estimated by measuring the vertical distance from
92

the ground to the lowest substantial density of live foliage on the trunk, guided by
hypsometer readings. These methods ensured that all measurements were not only precise
but also standardized across all plots, providing a high level of data quality.
Additionally, allometric equations, which relate easily measured tree characteristics
like species, DBH, and height to biomass estimates, were applied to calculate attributes such
as canopy bulk density (CBD), canopy fuel load (CFL), and vegetation density values across
three critical fuel layers: canopy fuel (CF), ladder fuel (LF), and understory fuel (UF). The
additional fuel layers, prioritized based on community input, were calculated using the
custom biomass-summing function and applied to the ALS-derived point cloud data, which
was binned into 1-meter height increments, as outlined in Chapter 2. Fuel loads were divided
into the three fuel categories and point density values were binned based on height:
understory fuel (0-2 meters), ladder fuel (2 meters up to the median canopy base height), and
canopy fuel (median canopy base height and above). Refer to Chapter 2 for more details on
the SE data collection.
Ocular Estimates (OE) Sampling
The OE method focused on rapidly collecting visual estimates of forest fuel densities
across the Xáxli’p Survival Territory. In contrast to SE sampling, which followed a stratified
random framework and involved detailed physical measurements, the OE method relied on
visual estimates of fuel loads, with sampling locations selected opportunistically based on the
local crew's deep knowledge of the landscape and its fuel distributions. Sampling locations
were chosen to represent a broad range of fuel densities, from low to high, while prioritizing
areas accessible within the constraints of time and resources. Unlike SE sampling, OE did not
include high-elevation and topographically challenging areas. This exclusion was due to
93

weather, time, and access constraints, as snow in higher elevations during the November
sampling period made these areas inaccessible. Instead, OE focused on familiar, easily
reachable locations. This approach leveraged the crew’s local knowledge and allowed for a
much larger number of samples to be collected compared to SE, thus providing broader
spatial coverage.
In October 2022, a preliminary dataset was collected in collaboration with the
community. This dataset comprised visual estimates of fuel loads for "Lower Fuel" and
"Canopy Fuel" levels, measured on a 1-3 scale (low, medium, high) for each fuel layer. This
dataset was not incorporated into the analysis due to a decision to refine the data collection
approach based on field learning from these initial OE samples. Key insights from this period
indicated that understory and ladder fuels should be recorded separately instead of combined
into a single "Lower Fuel" category. Additionally, the 1-3 scale was deemed insufficiently
nuanced, leading to the adoption of a 1-5 scale (low, medium-low, medium, medium-high,
high) in subsequent data collection efforts. This iteration also included enhancements in
group calibration and quality assurance processes.
Following the assessment and discussion of these October learnings, the community
reconvened in early November 2022, refining the data collection methodology. This revised
approach, which emphasized distinct vertical fuel profiles and a more nuanced 1-5 fuel load
scale, was then used to collect the nearly 700 samples for the OE method. This process
exemplifies a community-directed research approach, allowing for real-time learning,
engagement, and ownership over data collection and information generation. See Figure 30
for a map displaying the sample locations from all three sampling campaigns, highlighting
the spatial distribution and coverage achieved through the OE method.

94

Figure 30. Map of sample locations from all three sampling campaigns. This includes October and
November OE sample campaigns and SE method samples, highlighting spatial distribution and
coverage achieved.

To ensure consistency and accuracy in visual estimates, the crew underwent
comprehensive calibration exercises before data collection. The process began with an officebased review of maps of known high, medium, and lower-density fuel load areas, considering
reference photographs depicting a range of fuel densities for canopy, ladder, and understory
strata. This review provided an initial framework for alignment, supplemented by two days
of visual estimate calibration in the field throughout the Fountain Valley.
95

Field calibration involved visiting various well-known sites representing the full
spectrum of fuel densities within the Fountain Valley. These sites were collaboratively chosen
based on the crew’s extensive local knowledge, ensuring diverse fuel conditions were
thoroughly observed and assessed. At each site, crew members independently assessed the
fuel loads for each stratum and then participated in group discussions to reconcile differences
and reach consensus. This iterative, collaborative process emphasized relative fuel loads
specific to the Fountain Valley, ensuring that estimates were grounded in local environmental
conditions and informed by community knowledge. By fostering clear communication and
refining criteria for visual estimation, the calibration exercises aligned individual crew
members’ interpretations and established a unified approach for the data collection phase.
Sampling sites for the data collection campaign were identified using a combination
of local knowledge and consultations with satellite imagery and topographic base maps. The
crew targeted areas they believed represented high, medium, and low fuel densities across all
three vertical strata. This approach leveraged local expertise and knowledge to locate sites
that reflected the diverse fuel conditions of the territory while balancing systematic selection
with the flexibility of opportunistic sampling during routine fieldwork for other ongoing
projects. This starkly contrasts the systematic stratified random sampling framework used in
the SE method. While the OE method may be critiqued as less-scientific, not easily
repeatable, and not as robust, it represents an important experiment in incorporating local,
context-specific insights and knowledge into this type of work. This emphasis on local
expertise is a core part of the essence of the OE method.
At each OE plot, fuel density observations were recorded for the understory, ladder,
and canopy strata in a consistent sequence, beginning with the understory and moving

96

upward. A 1-to-5 scale (low to high) was used to quantify fuel densities, providing a
straightforward yet adequately nuanced framework for visual assessments. Plot sizes were
estimated visually to align as closely as possible with the 11.28-meter radius circular plots
used in SE sampling. Observations were made from the plot center, with crew members
visually scanning the surrounding area to estimate fuel loads. Sample data was recorded on
iPads and tagged to the GPS coordinates at the plot center.
The opportunistic nature of OE sampling enabled data collection during routine
fieldwork when crew members encountered sites representative of specific fuel densities. To
mitigate clustering and promote spatial diversity, the crew was trained to maintain a
minimum distance of 20 meters between sample plots and avoid taking multiple samples in
areas with high local variability. These guidelines ensured broader coverage while aiming to
minimize redundancy in highly variable sites.
The OE sampling approach was shaped by community priorities, emphasizing the
mapping of fuel layers most relevant to the Xaxli’p Community Forest wildfire risk
management. By leveraging local knowledge and focusing on areas deemed important and
representative by the crew, the method ensured that the collected data was directly applicable
to the community's needs. As local experts on wildfire risk and history in their Territory, the
crew possess deep knowledge of the territory and its fuel distributions. They are actively
involved in all aspects of fuel treatment and wildfire risk mitigation efforts. Moreover, they
have seen the history of wildfires burning within their Territory and the surrounding areas
over many years, giving them a tangible and application-focused interest in accessing forest
fuel maps that describe the conditions most relevant to their work.

97

The direct connection between data collection and practical wildfire management
highlights the importance of creating maps that align with on-the-ground realities and
community priorities (Baron et al., 2024). This collaborative and flexible approach also
highlights the value of incorporating local, context-specific insights into forest fuel mapping
(Aragoneses & Chuvieco, 2021; Engelstad et al., 2019). While the OE method lacks the rigid
structure and scientific repeatability of the SE approach, it compensates by enabling a
nuanced understanding of fuel distributions informed by firsthand experience and
engagement with the landscape. Additionally, it makes this type of field data collection far
more accessible at a small local community level compared to the more structured and
technically demanding SE method, which requires more staff time and resources to facilitate
travel to randomized stratified sample sites, take detailed measurements, and operate
specialized measurement equipment. These localized insights not only enriched the data but
also set the stage for integrating and normalizing the outputs of the SE and OE methods,
providing a foundation for effective cross-method comparisons in subsequent analysis.
Random Forest Modeling and Validation
As outlined in Chapter 2, ALS data was collected across the study area, and point
cloud metrics were calculated using 10-meter resolution rasters to cover the landscape
comprehensively. These metrics included height-binned point densities and other vegetation
structure attributes relevant to the prediction of fuel layers. For details on ALS data collection
and processing, refer to Chapter 2.
For the analysis in this chapter, both the empirical SE and OE datasets were used to
train Random Forest (RF) models to predict the three fuel classes of interest to the
community: canopy, ladder, and understory fuel. To ensure consistency, the modelling
98

procedures were identical for SE and OE model runs. These procedures included
hyperparameter tuning to optimize model performance, variable selection to identify the most
relevant ALS-derived metrics, and model validation to assess accuracy and robustness. This
uniform approach ensured that differences in model outcomes could be attributed to the
sampling techniques rather than inconsistencies in the modeling process.
A k-fold cross-validation technique was employed to evaluate model accuracy and
robustness for both SE and OE datasets. Key performance metrics included out-of-bag R²,
cross-validated R², and validated R², providing reliable measures of the predictive
performance for each dataset. These metrics were generated during the k-fold crossvalidation phase of model development, where the dataset was partitioned into subsets to
iteratively train and validate the model. The cross-validated R² values offer a useful
assessment of model performance by indicating how well the model can generalize its
predictions to unseen data based on the field samples that were systematically held out during
training. Nonetheless, these values are not definitive validation of the model’s accuracy in a
real-world scenario. Instead, these metrics represent one of several possible ways to evaluate
model performance, reflecting how well the model fits the available data under specific
controlled conditions. They offer insights into the relative strengths of SE and OE sampling
approaches. However, they do not capture all aspects of model reliability, such as uncertainty
in predictions, potential biases in the training data, or the complex variability found in natural
landscapes. While cross-validated R² values provide a helpful benchmark, they should be
interpreted with caution, acknowledging the limitations inherent in any modeling process.
All statistical analyses and model training were conducted using the R programming
language (R Core Team, 2023), ensuring a consistent analytical framework. By maintaining

99

uniformity in model configurations and using identical ALS-derived metrics for landscapelevel predictions, the study isolated the impact of sampling methods on model performance,
enabling a direct and meaningful comparison between the SE and OE datasets. For additional
details on the Random Forest modeling procedures, including specific parameters and
configurations, please refer to Chapter 2.
Analysis
The SE and OE samples were collected in different locations on the landscape, with
no two samples taken at the same spot. To compare and validate the models, we examined
how well the predicted values matched the observed values at their respective sample
locations. We conducted this comparison in two ways: within-model comparisons and crossmodel comparisons. This means that for the SE model, the predicted values were compared
to the observed SE sample data collected in the field. Additionally, the SE model’s predicted
values were compared to the observed OE sample data at the OE sample locations. This
allowed us to evaluate how well the SE model performed within its own sample set and when
it was validated against the observed values of fuel density collected at the OE sample
locations.
Similarly, for the OE model, we compared its predicted values to the observed OE
sample data and to the observed SE sample data. This approach ensured that both models
were tested against each other's sample locations, providing a unique and unconventional
validation process however with implicit inevitable limitations. While this approach may
differ from traditional methods, it allowed us to explore the strengths, weaknesses, and
relative locally validated accuracy of each method in a meaningful way. This process also
embodied the core ethic of the community-directed approach by involving the Xaxli’p
100

Community Forest crew in the validation process in the most tangible and direct manner. It
emphasized the importance of local, contextually relevant information and empowered the
community to engage with the generated layers while gaining a deeper understanding of their
strengths and limitations.
By assessing how well the initial SE layers predicted fuel concentrations based on OE
sampling data and vice versa, this approach honed in on the essence of the OE method:
integrating community-directed input and tying locally specific knowledge to the assessment
of the generated layers. This comparison was conducted for all three fuel layers of interest,
understory, ladder, and canopy fuels. By comparing predicted and observed values within and
across models, we were able to assess the accuracy and performance of both the SE and OE
methods comprehensively.
Both comparisons used scatter plots to depict the predicted vs. observed values for
each fuel type, Canopy Fuel (CF), Ladder Fuel (LF), and Understory Fuel (UF),within and
across the SE and OE predicted fuel layers. This enabled a direct comparison between SE
and OE sampling methods for each fuel type, providing an evaluation of the strengths and
weaknesses of each sampling method and offering a deeper understanding of their impact on
fuel layer predictions. Pearson’s correlation coefficient was calculated to quantify the
strength and direction of the relationship between predicted and observed values for each fuel
layer. The coefficient of determination (R²) was also calculated to assess the proportion of
variance in observed values that could be explained by the model’s predictions. These
metrics provided a standardized basis for evaluating model performance across SE and OE
datasets.

101

To visually compare the outputs from the SE and OE sampling methods, fuel load
maps for each of the three fuel layers, Canopy Fuel (CF), Ladder Fuel (LF), and Understory
Fuel (UF), were displayed side by side. This side-by-side display provided a clear visual
representation of the similarities and differences between the two methods, helping to
highlight where and to what degree fuel load predictions varied between each method across
the landscape. Difference maps were also calculated using raster subtraction. These were
displayed alongside the respective SE and OE fuel load maps to highlight areas where the
two sampling techniques differed.
Results
Sample Coverage Assessment
A comparison of the spatial coverage of SE and OE sampling techniques highlights
their differing strengths and limitations, as illustrated in Figure 31. The SE sampling strategy
involved 104 plots following a random stratified sampling design, ensuring representation
across Canadian Forest Fire Danger Rating System (FBP SYSTEM) fuel types and six broad
elevation bands. This approach included high-elevation plots, capturing a wider diversity of
forest types and fuel structures, including those unique to these areas. However, the timeintensive nature of SE sampling limited the total number of plots, restricting its overall
geographic coverage.
In contrast, the 695 OE samples were strategically distributed based on local
knowledge to represent different fuel densities throughout the territory. This opportunistic
sampling approach allowed for a larger number of samples to be collected in a shorter
timeframe. However, OE sampling was more concentrated in valley bottom areas near main
road access and did not achieve the same broad geographic extent as SE. The comparative
102

lack of coverage across the full territory was influenced by adjustments made following the
October sampling campaign, as high-elevation locations became inaccessible for the
November sampling due to snow. Despite this, the OE method was strategically designed to
leverage local knowledge of forest types, ensuring that sampling efforts targeted areas known
to be representative of a full range of fuel concentrations across the vertical strata. While this
approach enabled fine-scale spatial heterogeneity to be captured in accessible regions, it
resulted in less representation of fuel conditions in higher elevations and complex terrain
features such as gullies, ridges, and steep slopes.

103

Figure 31. Map showing the distribution of SE and OE sample locations within the Xáxli’p
Community Forest/Survival Territory. Red dots represent the 695 OE samples, and blue dots represent
the 104 SE samples.

104

Model Performance
The R² values in Table 3 highlight the performance differences between SE and OE models across the
three fuel layers.
Table 4. Cross-validated R² values for Random Forest models trained on SE and OE data for
predicting Canopy Fuel, Ladder Fuel, and Understory Fuel layers.

Response Variable

Structure Empirical

Ocular Estimate

Canopy Fuel (CF)

0.752

0.675

Ladder Fuel (LF)

0.559

0.621

Understory Fuel (GF)

0.292

0.486

These values represent the proportion of variance in observed data explained by each model,
highlighting the strengths of each sampling approach for different fuel layers.

The SE model achieved the highest R² value for canopy fuel at 0.752, demonstrating
strong predictive accuracy for this layer. Similarly, the OE model performed well for canopy
fuel with an R² value of 0.675, though slightly lower than SE. For ladder fuel, the OE model
surpassed SE with an R² of 0.621 compared to 0.559. In the understory fuel layer, the OE
model again outperformed SE with an R² of 0.486 versus SE’s 0.292. These values reflect the
distinct strengths of each sampling approach for different forest fuel strata.
Within-Model Comparison
The comparison of predicted versus observed fuel loads for canopy, ladder, and
understory layers reveals key differences in model performance between the structured
empirical (SE) and opportunistic ocular estimate (OE) methods. The SE dataset comprised
104 samples, while the OE dataset included 695 samples, approximately 6.7 times larger,
however many samples were tightly spaced together, therefore influencing spatial
dependence. This difference in sample sizes influenced the visualization and interpretation of
results.
For SE data, training and predicted values were continuous and normalized to a 0–1
scale. OE data were collected as discrete categories (1, 2, 3, 4, 5) and treated as a continuous
scale during Random Forest regression modeling. To visualize the results, scatterplots were
used for SE data, displaying the relationship between continuous predicted and observed
values. For OE data, violin plots were used to show the relative concentration of overlapping
points at each level.

105

Figure 32. Within-model comparison of predicted versus observed fuel loads for Canopy Fuel (top),
Ladder Fuel (middle), and Understory Fuel (bottom).
The comparison was done across Structured Empirical (SE) and Opportunistic Ocular Estimate (OE)
methods. For SE samples, red dots represent predicted values displayed using a traditional scatterplot
format. For OE samples, blue violin plots illustrate the relative concentration of overlapping values,
with individual OE points jittered for clarity. A faint dashed 1:1 line serves as a reference for perfect
prediction accuracy. Correlation coefficients for both SE and OE data are included in each plot,
providing a measure of the strength of the model's performance for each fuel layer. All fuel layer
metrics were normalized to a 0–1 scale to facilitate consistent comparison across models and layers.

106

Canopy fuel models produced higher accuracy for SE data, with a cross-validated R²
of 0.88 compared to 0.72 for OE models. The SE method exhibited a consistent tendency to
overpredict across all canopy fuel levels, as indicated by the general positioning of SE points
above the 1:1 line. In contrast, compared to the generally weak model for SE understory, OE
models overpredicted at lower observed canopy fuel levels and underpredicted at higher
levels, as seen in the violin plot distribution. Low observed values tended to be above the 1:1
line, while high observed values were generally below it.
Notably, both SE and OE models had relatively few samples in the highest canopy
fuel categories, with only a handful for OE and a single high-value sample for SE. The OE
model appeared to flatten predictions for high observed canopy fuel levels, rarely predicting
values beyond 0.75 even when observed canopy fuel reached 1.00. This suggests a limitation
in OE’s ability to capture extreme canopy fuel values. The violin plot for OE also shows a
higher density of predicted values around mid-range canopy fuel values (~0.5 - 0.75),
reinforcing its tendency to underpredict high canopy fuel loads while providing reasonable
estimates for mid-range conditions.
Ladder fuel models showed comparable performance for SE and OE datasets, with
cross-validated R² values of 0.70 and 0.69, respectively. Both models accurately identified
sites with an absence of ladder fuels but showed increasing variability at intermediate and
high ladder fuel levels. The OE method demonstrated better alignment with observed values
at mid-range ladder fuel concentrations (~0.25-0.50), as seen in the violin plot bulge centered
around the 1:1 line. Despite this, considerable variability remained in OE predictions across
all fuel levels. The SE model, on the other hand, exhibited a stronger tendency to overpredict
across most ladder fuel levels, with predicted values generally positioned above the 1:1 line.

107

Unlike the OE model, the SE model did not show improved accuracy at intermediate ladder
fuel values, with no clear increase in agreement at mid-range fuel levels. At the highest
ladder fuel levels (~1.00), the OE model again exhibited a flattening effect, with predicted
values clustering below the 1:1 line, similar to its behavior in the canopy fuel models.
Understory fuel models had the lowest predictive accuracy of the three fuel layers. SE
models achieved a cross-validated R² of 0.62, whereas OE models had an R² of 0.45. Neither
method showed strong alignment with the 1:1 line, indicating overall weaker predictive
performance. The OE model exhibited poor agreement at both low and high understory fuel
concentrations, with overprediction at lower levels (~0.00-0.25) and underprediction at
higher levels (~0.75-1.00). The violin plot and jittered points illustrate this trend, with low
observed values tending to be overpredicted (above the 1:1 line) and high observed values
underpredicted (below the 1:1 line).
Some moderate agreement was observed at middle to high fuel concentrations (~0.500.75) but with considerable spread, indicating inconsistent predictions in this range. For SE,
predictions were generally concentrated in the lower range of understory fuel values, with
most SE points distributed toward the lower end of the observed values. There were no highvalue SE samples for understory fuels, meaning that SE model performance at the highest
understory fuel levels could not be fully assessed. The single high-value SE sample appeared
to be underestimated, reinforcing the limitation of the SE method in capturing high
understory fuel conditions.
The OE models show a tendency to flatten predictions at the highest observed fuel
levels across all three categories, suggesting a limitation in their ability to capture extreme
values. SE models systematically overpredict across all categories, particularly in canopy and

108

ladder fuels. Understory fuel predictions show the highest variability, with neither method
demonstrating strong agreement with the 1:1 line. Mid-range ladder fuel estimates align
better in the OE model compared to SE, suggesting some advantage of the OE approach in
capturing variability in ladder fuels. These findings highlight the distinct performance trends
between SE and OE, particularly regarding overprediction tendencies in SE and constrained
prediction ranges in OE, which appear most pronounced in high fuel load conditions.
Cross-Model Comparison
We conducted cross-model comparisons for all three fuel layers to assess the
predictive capability of Structured Empirical (SE) and Ocular Estimate (OE) methods across
fuel types. The OE samples are based on the visual estimates collected in the field, whereas
the observed values for the SE sample plot locations are ALS-derived values based on point
cloud density and distribution.
For each plot (see Figure 33), predicted values from SE models are compared to
observed values from OE sample locations (left column), and predicted values from OE
models are compared to observed SE values (right column). In these plots, the black line
represents the regression line of the predicted fuel layer values versus the observed amount of
that fuel layer estimated using the other sampling method. Specifically, the left column
compares predicted fuel loads from the SE-trained model against observed fuel loads from
empirical OE sampling, and the right column compares predicted fuel amounts from the OEtrained model against observed fuel loads from empirical SE sampling.

109

Figure 33. This figure displays cross-model comparisons for Canopy Fuel (top), Ladder Fuel
(middle), and Understory Fuel (bottom) layers. In each plot, the left side shows a violin plot with
individual values represented as jittered blue dots for the predicted SE values against observed OE
values, while the right side shows predicted OE values against ALS-derived observed SE values. A
red dashed 1:1 line serves as a reference for perfect prediction accuracy, with a solid black line of best
fit illustrating the trend in the data. Pearson’s correlation coefficients and R² values indicate the
predictive performance for each fuel type across both methods.

110

The closer the black regression line is to the slope of the red one-to-one line, the
better the model captures the variance in fuel loading across the landscape. However, it is
important to note that a model exhibiting the same slope as the red one-to-one line (i.e., the
black and red lines align) may still reflect low model accuracy if residual variation is high.
High residual variation indicates notable deviations between predicted and observed values,
reducing the reliability of the model's predictions despite the apparent alignment of slopes.
For canopy fuel, the SE model demonstrated slightly better predictive capability when
applied to OE sample locations, with a correlation of 0.571 and an R² of 0.326 compared to
the OE model, which achieved a correlation of 0.54 and an R² of 0.292. While both models
captured the general trend of canopy fuel density, deviations from the 1:1 line were observed,
particularly at higher observed canopy fuel values for both models and at lower observed
values for the OE model. Both methods tended to underpredict high canopy fuel values, and
the OE model overpredicted at lower fuel density values. Observations of high canopy fuel
loads were limited, which likely contributed to the poor fit at these densities.
For ladder fuel, predictive performance was weaker than for canopy fuel. When used
to predict OE-observed ladder fuel, the SE model achieved a correlation of 0.328 and an R²
of 0.108. The OE model applied to ALS-derived SE-observed ladder fuel yielded a
correlation of 0.316 and an R² of 0.1. Scatterplots for both models showed high variability,
with points diverging from the 1:1 line and reflecting inconsistent model performance. The
SE and OE models faced challenges in accurately predicting ladder fuels due to their
inherently discontinuous and variable structure.
For understory fuel, cross-model predictive performance was the poorest among the
three layers. The SE model predicting OE-observed understory fuel yielded a negative

111

correlation of -0.112 and an R² of 0.013, while the OE model predicting ALS-derived SEobserved understory fuel showed a near-zero correlation of -0.004 and an R² of 0. Both
models exhibited challenges in capturing understory fuel variability when cross-compared, as
evidenced by scatterplots showing flat regression lines and a wide spread of points. The
clustering of ALS-derived SE-observed values near zero suggested that the SE method
underrepresents variability in understory fuels.
Across all three fuel layers, canopy fuel predictions showed the highest levels of
agreement, while ladder and understory fuel predictions exhibited weaker correlations. Both
SE and OE methods struggled to model intermediate and high variability fuel layers,
reflecting the inherent challenges of modeling complex and heterogeneous vegetation
structures.
Fuel Maps Comparison
This section presents the differences between structured empirical (SE) and ocular
estimate (OE) methods by comparing fuel load prediction maps across Canopy Fuel (CF),
Ladder Fuel (LF), and Understory Fuel (UF) layers (Figure 34). Difference maps were
created by subtracting the OE-based fuel load predictions from the SE-based predictions for
each fuel type (positive values correspond to OE-predicted fuel values being higher than SE,
while negative values indicate where OE predicts lower values). These maps highlight areas
where the methods differ most substantially, providing insights into the spatial variability of
discrepancies.
The difference maps help identify regions with the greatest divergences between SE
and OE methods, helping to understand how each method captures fuel load density across
the landscape.
112

Figure 34. Comparison of Canopy, Ladder, and Understory Fuel Layers and Their Differences Between Structured
Empirical and Ocular Estimates. This figure displays fuel load predictions for Canopy Fuel (top), Ladder Fuel
(middle), and Understory Fuel (bottom), with maps derived from SE data (left), OE data (center), and a difference
map (right) for each fuel layer. The difference maps highlight areas where SE and OE estimates diverge, with blue
indicating areas where SE predictions are higher than OE, turquoise representing moderate negative differences
(OE lower than SE), black showing no difference between methods (OE ≈ SE), orange indicating moderate
positive differences (OE higher than SE), and red highlighting areas where OE predictions are higher than SE.
This consistent color scheme emphasizes spatial patterns and discrepancies between the two methods.

113

At a landscape level, the SE and OE methods effectively identify broad patterns of
fuel load distribution, particularly in low-density fuel areas such as alpine regions with
minimal forest cover and open grassy areas along riverbanks in the northwestern portion of
the study area. However, discrepancies between SE and OE predictions are evident in high
and medium-density fuel areas, where SE tends to predict higher values. These discrepancies
are most pronounced in understory fuel layers, where SE generally predicts lower fuel
densities than OE across the study area, with notable differences in specific regions, such as
cut blocks in the northeast.
Discussion
The comparison between the structured empirical (SE) and opportunistic ocular
estimates (OE) methods for forest fuel mapping is presented here. The focus is on sample
coverage assessment, model performance, comparisons of predicted vs. observed values, and
spatial analysis. This comparison of the two distinct sampling techniques provides a nuanced
understanding of the benefits and trade-offs of each, highlighting their impact on predicting
critical fuel layers and informing strategies for future community-led wildfire risk
management initiatives.
Sample Coverage Assessment
The contrasting spatial coverage of SE and OE sampling methods highlights key
trade-offs between systematic design and opportunistic, knowledge-driven approaches. SE's
stratified random framework ensured broad representation across varied fuel types and
elevation bands, particularly capturing critical high-elevation areas with unique forest
characteristics. This systematic approach contributes to structural diversity representation and
robustness in modeling. However, the limited number of SE plots underscores its primary
114

limitation, reduced geographic coverage due to the intensive time and resource demands of
the method. Despite this, SE remains the more thorough sampling approach. In contrast, the
OE method appears to have a larger effective sample size, but this is influenced by spatial
dependence, which can artificially inflate its coverage.
Conversely, the OE method’s reliance on local expertise allowed for far greater
spatial coverage, enabling the collection of nearly 700 samples within a much shorter
timeframe. This method’s ability to focus on fine-scale variability within accessible areas
highlights the value of community-driven sampling. However, its exclusion of high-elevation
areas and complex terrain introduces gaps in the representation of landscape-wide fuel
diversity. The reliance on accessible, familiar locations may also lead to a bias toward areas
with more moderate conditions, potentially missing critical variations in challenging regions.
These differences in spatial coverage have notable implications for the accuracy and
reliability of models derived from each sampling method. SE sampling provides robust data
for stratified, systematic analysis but lacks the flexibility, local context, and implicit
knowledge embedded in the OE sampling. Ultimately, this analysis underscores the
complementary nature of these methods. While sample coverage assessment is a foundational
step in understanding the differences between SE and OE methods, it represents just one of
many dimensions explored in this study.
Moreover, as Spoon (2014) highlighted in his work on Indigenous ecological
knowledge, the iterative refinement of OE sampling demonstrates the adaptability of this
approach within a community-driven context. Challenges faced during the October
campaign, including sampling inconsistencies, provided valuable lessons that informed the
improved November protocols. This iterative learning aligns with the ethos of community-

115

based research by incorporating local knowledge and fostering ongoing refinement. It also
emphasizes the importance of calibration, training, and clear protocols in ensuring consistent
and accurate data collection (Arcidiacono et al., 2014).
The treatment of OE data as continuous for modeling and analysis purposes
introduces both strengths and limitations to the study. By assuming equal intervals between
ordinal categories, the approach aligns with the intention of approximating a continuum of
fuel density during field data collection. This assumption facilitated direct comparisons
between SE and OE models, allowing for unified analysis and methodological consistency.
However, this decision also carries potential drawbacks, particularly in the context of
correlation testing.
Pearson’s correlation, which assumes continuous data, may have been differentially
sensitive to the ordinal nature of OE data. The discrete categories may not fully capture the
underlying variability of fuel loads, leading to potential overestimation or underestimation of
the strength of the relationship between predicted and observed values. While Random Forest
regression is robust to this assumption, treating ordinal data as continuous could influence the
model’s ability to detect finer-scale patterns or variability within the OE dataset, however this
is a fairly common approach for this type of analysis.
Despite these limitations, the decision to treat OE data as continuous is justified,
given the practical constraints of field sampling and the need to balance precision and
scalability in the analysis. This approach reflects the study’s emphasis on leveraging local
expertise and rapid, large-scale data collection. As Arcidiacono et al. (2014) discussed, the
transition between qualitative and quantitative methods can provide advantages in
understanding complex ecological patterns. Future studies could explore alternative methods,

116

such as ordinal regression or rank-based approaches, to better assess the sensitivity of models
to ordinal data and to validate the robustness of this assumption in different contexts.
Model Performance
The results demonstrate the unique advantages and limitations of the SE and OE
sampling methods for predicting forest fuel layers. SE’s structured sampling approach excels
in predicting canopy fuel, likely due to the inclusion of precise physical measurements of
attributes like tree height, diameter, and crown structure. This systematic method ensures
accuracy for the upper layers of the forest but is limited in its ability to capture variability in
ladder and understory fuels.
In contrast, the OE method’s flexibility and larger sample size contributed to its
stronger performance in predicting ladder and understory fuels. These strata are inherently
more variable and discontinuous, which the opportunistic sampling approach was better
equipped to address. Using localized knowledge allowed the OE method to target areas with
fine-scale heterogeneity, resulting in higher R² values for these layers. However, compared to
SE, the reliance of OE on visual estimates and the absence of precise physical measurements
may have slightly reduced its predictive accuracy for canopy fuels.
Together, these findings emphasize the need for tailoring sampling strategies to the
specific characteristics of each fuel layer. SE is well-suited for applications requiring detailed
canopy fuel predictions, while OE offers an effective approach for capturing variability in
lower strata. Integrating the strengths of both methods could enhance future modeling efforts,
enabling more comprehensive and accurate wildfire risk assessments.
As mentioned earlier, another important factor to consider is the distribution of
sample locations between the two approaches. The SE technique likely captured a broader

117

range of forest structural characteristics and topographical variation across the study area,
particularly because it included samples in higher elevation areas where the OE method did
not. These high-elevation areas are characterized by notably different forest types, which can
have strong implications for the predictive power of the models. The absence of OE samples
in these areas may have limited its ability to generalize across the entire landscape,
potentially impacting its overall performance at the landscape level.
Estimating canopy fuel load from the ground also poses an inherent challenge for the
OE approach, particularly due to visibility issues caused by lower vegetation layers
obstructing the view of the canopy. Despite these challenges, the OE model still performed
well, highlighting that even with certain limitations, visual estimates can offer a viable
method for capturing canopy fuels. This underscores that while structured, precise
measurements are valuable for capturing the finer details of the forest structure, opportunistic
visual assessments can still provide a reasonably accurate representation, making OE a
practical option, particularly when resources are constrained or when a rapid, accessible
sampling approach is needed. However, it is important to acknowledge that ALS best
practices for field data collection align more closely with SE than OE. ALS-based modeling
requires high-quality, well-structured ground data for calibration and validation. SE's
systematic approach, with precise and repeatable measurements, ensures that field data can
be directly linked to ALS-derived metrics, improving model accuracy and reliability. In
contrast, OE's more subjective and variable nature makes it less compatible with ALS
requirements, as visual estimates may introduce inconsistencies that reduce their
effectiveness as training data.

118

This distinction highlights a key consideration when integrating ALS with field data
collection: while OE may provide broader spatial coverage and accessibility, SE is better
suited for producing high-quality reference data that aligns with ALS-derived variables.
OE provided stronger predictions for both ladder and understory fuels, with R² values
of 0.621 and 0.486, respectively, compared to SE’s 0.559 for ladder fuel and 0.292 for
understory fuel. The greater number of samples, strategic flexibility in selecting
representative sample sites, and OE's focused visual estimation approach enabled better
capture of variability in these lower strata. Ladder fuels, which are critical for understanding
vertical fire spread, were more effectively captured by OE’s opportunistic sampling, which
provided a more comprehensive dataset that was more representative of these layers than
SE’s more rigid, stratified framework. The adaptability of visual estimation methods allowed
OE to capture a wide range of ladder fuel densities, which are often challenging to measure
explicitly through the systematic physical measurements characteristic of SE.
For understory fuels, OE's advantage was the most pronounced. The localized,
complex nature of understory vegetation benefits from a rapid, adaptable sampling approach
that allows for a larger number of plots to be surveyed. It also benefits from incorporating
community knowledge into the sampling process to enable honing in on specific areas known
to have high, medium, and low fuel understory fuel density. This flexibility enabled OE to
capture spatial variability in understory fuels that SE’s more constrained approach may have
overlooked. The higher R² value for OE in predicting understory fuel emphasizes the value of
incorporating local expertise and adaptive sampling methods when dealing with vegetation
that exhibits high spatial variability close to the forest floor.

119

The comparison between SE and OE methods reveals key trade-offs. SE is wellsuited for providing detailed, precise measurements of canopy fuels, thanks to its structured
approach and the tools used to measure specific physical characteristics. However, the OE
method's much higher sample size and its particular effectiveness in capturing the variability
and complexity of ladder and understory fuels highlight its flexibility and the advantages of a
flexible, community-driven approach. Moreover, the lack of OE sampling in higher
elevations might have constrained its overall predictive capability for canopy fuels, while
SE’s inclusion of these diverse areas enhanced its robustness in these higher strata. Capturing
OE samples in the high elevation has the potential to change these results substantially.
Within-Model Comparison
The SE model demonstrated a higher correlation (0.88) compared to the OE model
(0.72) for canopy fuel, likely due to SE’s reliance on precise physical measurements such as
tree height, diameter, and crown base height, which are critical for accurately predicting
canopy fuel loads. However, the predicted values for SE samples appear to systematically
overestimate observed values, as the predicted values are consistently higher than the
observed ones across the range. While this overestimation might initially seem like a
limitation, it is arguably less problematic than a model that alternates between over and
underestimations, which could introduce more interpretive challenges.
In contrast, the broader spread and variability observed in the OE results, particularly
at higher observed canopy fuel values, reflect the inherent challenges of visually estimating
canopy fuels from the ground. These challenges are exacerbated by obstruction from lower
vegetation, making the upper layers more difficult to assess. Despite these limitations, the OE

120

model’s reasonable correlation (0.72) highlights its value for rapid, large-scale data
collection, providing useful approximations of canopy fuel loads.
For ladder fuel, the near-equal correlations of SE (0.70) and OE (0.69) models
suggest that both methods faced similar challenges in predicting this fuel layer. While SE
benefited from the precision of systematic physical measurements, it also exhibited
overestimation tendencies like those seen in the canopy fuel predictions.
For understory fuel, the SE model achieved a higher correlation (0.62) compared to
the OE model (0.45), but its sampling framework could have contributed to skewed results
toward the low (zero) understory fuel values. The SE data showed a consistent bias toward
lower observed values, likely due to the stratified sampling framework, which may have
coincidentally included areas with naturally low or absent understory fuels. Furthermore,
SE’s lack of robust field data collected for understory fuels at its plots suggests these results
may not fully capture the variability of this layer. The SE method did not collect explicit
measurements to describe understory fuels specifically in its fuel variable calculations. In
contrast, the OE method avoided this concentration at the low end, aiming for broader
representation across the full range of understory fuel conditions. However, OE
systematically overpredicted understory fuel loads and struggled to differentiate between
intermediate and high understory fuel values, which further underscores the difficulty of
modeling this highly variable layer.
Among the three layers, understory fuel stands out as particularly challenging to
predict, as reflected in the lower correlations for both methods. This result underscores the
inherent complexity and heterogeneity of understory vegetation, which is influenced by
highly localized factors that are difficult to capture with either sampling method (García-

121

Cimarras et al., 2021). These findings emphasize the importance of understanding how
sampling frameworks influence model performance.
SE’s systematic and precise measurements provided higher accuracy for canopy and
understory fuels but struggled to represent variability in more heterogeneous layers, such as
ladder and understory fuels, due to its limited sample size. While less precise in its individual
predictions, OE excelled in capturing broader landscape variability, which is critical for
large-scale assessments. The consistent overestimation by SE and OE models across several
layers might be less problematic than inconsistent patterns of over and underestimation,
which could complicate the interpretation and application of the results.
Cross-Model Comparison
The results highlight the differing strengths and limitations of SE and OE methods for
cross-model predictions. For canopy fuel, the SE model’s higher correlation and R² values
underscore its effectiveness in capturing structural attributes such as tree height, diameter,
and crown base height. However, the consistent overprediction of canopy fuel values by both
models suggests potential biases in sampling frameworks or ALS-derived observed values.
These systematic overpredictions are less problematic than alternating over and
underpredictions, which could introduce greater uncertainty in practical applications. The SE
method appears to be more consistent in its overprediction across all fuel levels, whereas the
OE method exhibits a more variable pattern, tending to overpredict at lower fuel values and
underpredict at higher fuel values. This pendulum effect in OE predictions introduces
additional challenges in interpretation and application, as it results in greater fluctuations in
error across different fuel densities as well as practical interpretability. The tendency of the
OE method to systematically shift between over and underprediction, rather than maintaining

122

a consistent bias, makes it more difficult to correct and increases uncertainty in its use for
predictive modeling and wildfire fuel assessments.
The similar performance of SE and OE models for ladder fuel reflects shared
challenges in modeling this intermediate and discontinuous layer. Ladder fuels are inherently
variable, and neither method could accurately predict fuel loads consistently. The SE model’s
systematic approach offered precision, while the OE model’s larger sample size captured
broader variability. However, both methods exhibited wide spreads of predictions, suggesting
that additional refinement in sampling frameworks or modeling techniques is necessary to
improve accuracy. As with canopy fuel, the SE method tended to overpredict across all ladder
fuel densities, while the OE method followed a pattern of overpredicting at lower ladder fuel
levels and underpredicting at higher levels. This alternating prediction bias in OE results,
while introducing greater variability, also suggests that OE may better reflect the complexity
of ladder fuel distribution, particularly in capturing a wider range of fuel densities across the
landscape.
Understory fuels emerged as the most challenging layer to model, with near-zero
correlations for both methods in cross-model comparisons. The SE method’s stratified
framework and reliance on ALS-derived observed data likely contributed to the
underrepresentation of understory fuel variability, as observed values were heavily clustered
near zero. Conversely, while capturing broader representation, the OE method consistently
overpredicted fuel loads and struggled to differentiate intermediate and high understory fuel
levels. The pattern of overprediction at lower levels and underprediction at higher levels in
OE models was also apparent in understory fuel predictions, reinforcing the trend seen in
canopy and ladder fuels. These results reflect the highly localized and heterogeneous nature

123

of understory fuels, which are influenced by micro-environmental factors that neither
sampling method fully addresses (García-Cimarras et al., 2021). The lack of high-value
understory samples in SE data further constrained its ability to predict upper-range fuel loads,
while OE’s inconsistencies at different density levels added further uncertainty to its
predictive capacity.
The cross-model comparison emphasizes the influence of sampling frameworks on
model performance. SE’s structured empirical protocols captured detailed measurements for
canopy fuels but lacked flexibility for more variable layers like ladder and understory fuels.
OE’s adaptability and larger dataset provided a better representation of variability in ladder
and understory fuels but introduced biases and inconsistencies in canopy fuel predictions due
to reliance on visual estimation techniques. The tendency of OE models to alternate between
over and underprediction, rather than maintaining a stable bias, complicates their
interpretation and practical application. These trade-offs suggest that neither method is
wholly superior, and their strengths are complementary. Integrating the precision of SE for
canopy attributes with the scalability and adaptability of OE for variable layers could provide
a more robust approach to modeling forest fuel layers.
Fuel Maps Comparison
While SE and OE methods produce broadly similar landscape-scale predictions for
canopy and ladder fuels, important differences emerge at finer scales. SE tends to produce
greater spatial distinction between adjacent fuel densities, aligning well with topographical
features such as gullies, ridgelines, and drainage areas. This likely reflects the systematic
nature of SE’s random stratified sampling framework, which ensured a well-distributed
representation across diverse fuel types and elevation bands. In contrast, while effective for

124

general fuel density categories, OE’s reliance on local knowledge and opportunistic sampling
may have avoided challenging areas like steep slopes and drainage bottoms, reducing its
ability to capture site-specific topographical variations.
The OE-derived maps appear more diffuse and less aligned to topographical patterns,
especially in high-density fuel areas. While OE captures broad differences between high and
low fuel densities, it lacks the granularity and differentiation within high-density regions that
SE achieves. This is likely due to OE’s fixed 1–5 scale for fuel density, which limits its
ability to capture subtle differences. On the other hand, SE used continuous, precise
measurements, allowing for greater granularity and more detailed differentiation in fuel
density assessments.
The most substantial discrepancies occur in the understory fuel layer. SE predicts
much lower understory fuel densities overall, while OE predicts much higher values.
Difference maps highlight notable regions, such as some previously harvested and replanted
areas in the higher elevation areas of the northeastern parts of the Territory, where SE
predicts higher understory fuels. This may reflect SE’s ability to capture the dense, bushy
nature of young trees in replanted areas. However, OE’s sampling did not include these highelevation regions, nor did it clearly capture and distinguish between the harvested and
unharvested areas, potentially leading to underrepresentation in these forest structural types.
Conversely, SE’s stratified framework may have coincidentally targeted areas with naturally
low or absent understory fuels, limiting its ability to represent variability in this layer.
These findings outline the strengths and limitations of each method. SE’s ability to
align fuel density predictions with topography and to capture finer distinctions in medium-to
high-density areas suggests that it may provide more reliable data for detailed fuel mapping.

125

These distinctions are particularly critical for informing wildfire risk management in areas
like replanted cut blocks and steep terrains, where fuel density can vary substantially.
However, SE’s limited sample size and higher resource demands constrain its scalability for
larger landscape-wide assessments. In contrast, OE offers larger sample size and scalability,
making it more practical for large-scale assessments. Its flexibility allows for rapid data
collection, which is valuable for capturing variability in ladder and understory fuels.
However, OE’s fixed scale, reliance on visual estimation, and opportunistic nature limit its
ability to capture finer details or align fuel densities with specific topographical features.
An integrated approach combining SE’s precision with OE’s scalability could address
these limitations. For example, incorporating a stratified sampling strategy into OE’s
framework could improve its ability to detect subtle differences in high and medium-density
areas. Conversely, SE’s systematic methods could benefit from incorporating local
knowledge to expand its coverage and representation of diverse fuel conditions. Together,
these methods could provide forest managers with more accurate and comprehensive data for
wildfire risk mitigation and fuel management strategies.
Conclusion
This study compared structured empirical (SE) and opportunistic ocular estimate
(OE) methods for forest fuel mapping, highlighting their distinct advantages and limitations
(Table 4). SE provided high precision, particularly for canopy fuels, through its structured,
stratified sampling but required significant time, technical expertise, and resources. OE, on
the other hand, enabled rapid data collection, relied on local knowledge, and better captured
variability in ladder and understory fuels but introduced greater potential for bias and lower
accuracy in canopy estimates.
126

Table 5. Comparison of Structured Empirical (SE) and Opportunistic Ocular Estimate (OE) Methods
for Forest Fuel Mapping

Criteria
Cost
Sample Size
Spatial Coverage
Planning
Requirements

Technical Expertise
Precision
Criteria
Representation of
Ladder and
Understory Fuels
Alignment with
Topography

Bias Potential
Applicability to
Community-Led
Projects

Structured Empirical (SE)

Ocular Estimate (OE)
Low, requiring minimal tools
High due to specialized equipment, time, and resources, and much
and technical expertise required.
faster.
Fast and flexible allowing for
Limited due to time-intensive sampling
larger sample sizes in shorter
and strict plot locations.
timeframes.
Tendency to be concentrated
Broad due to random stratified
in easy to access areas and
framework.
clustered.
Minimal pre-planning; relies
Requires detailed pre-sampling planning on local knowledge for site
with a stratified framework.
selection.
Minimal technical expertise
Requires technical training and expertise needed; depends on local
to operate equipment and perform
familiarity with the
specific detailed measurements.
landscape.
Moderate; more variability in
High precision, particularly for canopy
estimates, particularly for
fuel measurements.
canopy fuels.
Structured Empirical (SE)
Ocular Estimate (OE)
Limited due to time intensive nature;
struggles with variability in these layers Stronger; visual estimate
and difficulty defining what
approach captures variability
characteristics to measure from these
in ladder and understory fuels
levels.
with greater nuance.
Weaker; less differentiation
Strong alignment; captures site-specific within topographically
variations tied to terrain features.
complex areas.
Potential for underrepresentation of
ladder and understory fuels as stratified
Potential for bias in canopy
sampling does not account for these
fuel estimates due to reliance
features.
on visual estimation.
Highly practical for
community projects with
Less practical for resource-limited,
limited resources and time
community-driven initiatives.
constraints.

These results illustrate a key trade-off: SE offers greater accuracy but at a higher
resource cost, while OE is more scalable and accessible but prone to inconsistencies. SE’s
127

tendency to overpredict canopy fuels is systematic and easier to adjust, whereas OE’s
fluctuations, overestimating low fuel values and underestimating high ones, introduce
challenges for bias correction. Despite this, OE remains a practical tool for broad-scale
assessments, particularly in resource-limited settings.
A hybrid approach is recommended, combining SE’s precision for canopy fuel
measurement with OE’s efficiency in capturing ladder and understory fuels. Integrating
elements of SE’s stratified sampling could improve OE’s landscape representation, while
selectively applying SE measurements in targeted plots would enhance accuracy without
overwhelming community resources. Standardizing OE visual estimates using reference
photos from locally relevant forests could further reduce variability and improve consistency.
This balanced approach would strengthen community-driven wildfire risk
assessments, making fuel mapping both precise and scalable. By improving accessibility and
leveraging local knowledge, communities can enhance their capacity to manage wildfire risks
without relying solely on intensive technical expertise.
In Chapter 4, we synthesize the broader outcomes of this study, exploring how the
integration of SE and OE methods, alongside community collaboration, led to the
development of a decision-support mapping tool that enhances the Xáxli’p community’s
ability to manage wildfire risks effectively.

128

Chapter 4: Integrating Local Knowledge and Technology: The Development of a
Community-Based Mapping Tool for Xáxli'p Land Stewardship

Figure 35. View of a switchback on Highway 99 and the east end of Seton Lake, just west of Lillooet.

The Evolution of a Community-Directed Research Partnership
This chapter blends third-person analysis with first-person reflection. The thirdperson sections provides objective discussion, while the first-person parts reflect personal
experiences and the collaborative nature of this work. This mix is intentional, aligning with
both the academic and community-driven aspects of the research partnership, which began as

129

a journey into community-directed research and an exploration of the Xáxli'p way of life,
their values, and their land stewardship practices.
At its core, this research partnership aimed to integrate the precision of high-density
ALS (LiDAR) data with local context and community knowledge to map forest fuels in a
way that was scientifically robust, culturally relevant, and ultimately useful for the Xáxli'p
community. Initially framed as a wildfire risk reduction and climate adaptation effort for the
Xáxli'p Survival Territory, the project evolved into a collaborative process of knowledge
integration, cultural continuity, and capacity-building.
From the outset, this research was driven by the need to ensure that high-resolution
forest fuel mapping could be accessible and useful within the Xáxli'p context. Beyond
technical accuracy, the challenge lay in bridging empirical scientific methods with traditional
and community knowledge and finding ways to make advanced modeling techniques
accessible and relevant. This required critically examining how different approaches to field
data collection, one structured and scientific, the other rooted in intuitive local expertise,
shaped the outcomes of ALS-driven forest fuel modeling. It also meant ensuring that the
resulting maps and data could be delivered in an intuitive format, one that would not only
support wildfire and land management but also serve as a tool for long-term community
engagement, planning, and knowledge sharing.
This thesis follows the journey of that process, examining the tensions and synergies
between technology and tradition, structured models and lived experience, and scientific
precision and local wisdom. More than just a study of fire risk mapping, it is an exploration
of how information is gathered, represented, and mobilized in ways that reflect the priorities
and self-determination of the Xáxli'p people.

130

Figure 36. Xáxli'p Community Forest Crew Leader, taking a measurement during field sampling on a
remote hillside. Xáxli'p Survival Territory, with Fountain Peak in the background.

This work was never intended to serve the priorities of academia or external agencies.
Instead, it was grounded in the desires and values of the Xáxli'p people and built upon
meaningful relationships, with members of the community like Ed, Rob, Derek, Karen, Puq,
Josiah, Leah, the entire forest crew, and the wider community. Together, we aimed to create
an enduring legacy of useful and relevant information from the ALS data, that would
empower the Xáxli'p people to protect and manage their lands far into the future. At its heart,
this was a project about stories, those held in the land, told by its people, and carried forward
through tradition and technology.

131

Figure 37. Researcher Patrick Robinson measuring tree height with a hypsometer while holding an
iPad for data entry. This was during the fuel sampling campaign in the Xáxli'p Survival Territory. The
yellow measuring tape marks the plot diameter.

As detailed in previous chapters, the project’s progression was methodical, adaptive,
and transformative. Chapter 1 explored the origins and evolution of the community-directed
partnership, delving into the history and ongoing journey of Xáxli'p land management.
Chapter 2 presented the technical analysis, focusing on the application of random forest
modeling to predict forest fuel types using ALS and field data. Chapter 3 compared two field
data collection techniques, examining how different training datasets influenced the
modelling results and considering the strengths and limitations of each method. This final
chapter brings these elements together, reflecting on the journey as a whole and highlighting
the development of a web-based mapping tool that serves as a culmination of this phase of
the project and a starting point for future applications.

132

Figure 38. Aerial view of the transition between a fuel-treated area (left) and a non-fuel-treated area
(right) of the forest near Kwotlenemo Lake.

As this project revealed, maps are more than just tools; they are representations of
cultural knowledge, connections between past and present, and guides for the future. They
carry the stories of the Xáxli'p, the wisdom of elders, the resilience of the community, and the
enduring spirit of the land. By unearthing old maps, data, and stories and making them
accessible and interactive, we worked cooperatively to enable the continuation of Xáxli'p
land management using the maps to support a renewed sense of clarity and agency. The
development of the web mapping platform represents a synthesis of traditional knowledge,
technical innovation, and practical utility, ensuring that the stories contained in these maps
remain vibrant and impactful for generations to come.
Importantly, these historical maps and records provide insight into how vegetation
patterns have changed over time. Historically, the landscape was a natural mosaic of different
forest types and fuel densities, shaped by both ecological processes and intentional fire
stewardship by the Xáxli'p. This region is naturally a fire-prone forest ecosystem with a
relatively frequent fire return interval. In the past, Xáxli'p fire-keeping practices, combined
with the absence of modern fire suppression, allowed for frequent low-severity fires that
133

maintained open stand structures, reduced understory density, and limited ladder fuels. These
fires helped keep fuel connectivity fragmented, creating a more fire-resilient landscape. In
contrast, today’s forests, shaped by decades of fire exclusion and modern land management,
have become denser, with continuous fuel connectivity, heavy ladder fuels, and increased
overall fuel load conditions that contribute to higher wildfire severity.
However, it is important to acknowledge that the available maps only go back so far,
and some of the information they contain is inferred rather than directly recorded. While they
do not provide a complete picture of historical vegetation states, they add another piece to the
puzzle, offering valuable evidence that helps to reconstruct past forest conditions. When
combined with oral histories, ecological knowledge, and other historical records, these maps
serve as an important tool for understanding landscape changes and guiding future land
stewardship efforts.

134

Figure 39. View looking north up the Fraser River Valley from a large, old burned area in the
northeastern portion of the Xáxli'p Survival Territory.

Key Findings
By merging modern mapping technology with community knowledge, this work has
demonstrated how collaborative efforts can produce practical, lasting solutions for land
stewardship and wildfire management. One of the major findings was the effectiveness of
combining Airborne Laser Scanning (ALS) with less structured, visually
estimated community-led data collection methods to create high-resolution, actionable
information on forest fuel layers, as described in Chapter 3. The data generated through this
work, encompassing canopy, ladder, and understory fuels, as well as other important forest
metrics, have provided the Xáxli'p with meaningful new insights into wildfire risk across
their territory. These findings provide tools for immediate risk mitigation and enduring
resources for long-term eco-cultural restoration planning.

135

Figure 40. Xáxli'p Community Forest crew leaders gathered around the web mapping tool in the
Xáxli'p Community Forest office during a fieldwork planning session.

While structured empirical (SE) methods offered high accuracy, ocular estimates
(OE) prioritized practicality, scalability, and community engagement. This balance ensured
that the project remained locally calibrated and accessible, empowering the community to
take ownership of data collection while integrating locally relevant knowledge into the
dataset. This not only contributed to the final wildfire fuel maps but also fostered ongoing,
independent data collection for future use. This adaptability reflects a sustainable approach to
community-based science, bridging the gap between academic rigor and local applicability.
Similar to the findings of Spoon (2014), incorporating local expertise into data collection can
enhance methodological flexibility and long-term engagement, reinforcing the value of
participatory research. Additionally, as Arcidiacono et al. (2014) stated, balancing structured

136

and flexible methodologies can strengthen both precision and contextual applicability,
making research outcomes more useful across diverse environments and cultures.

Figure 41. Field data collection equipment used for the structured empirical (SE) data collection. The
equipment includes a drone for airborne imaging, navigation, and situational awareness; a hypsometer
for measuring tree heights; a differential GPS unit for accurate plot center locations; a measuring tape
for determining plot diameters; an iPad with EpiCollect for streamlined data collection; and a
diameter tape for measuring tree DBH within plots.

The research partnership has highlighted the value of combining historical and newly
generated data into an integrated and cohesive platform. By consolidating legacy spatial data
with the new ALS-derived fuel maps and adding more publicly available spatial data, the
web mapping tool has transformed once fragmented information into an accessible and
adaptive resource. This integration ensures that previously overlooked or less accessible data
have become central to the community's land management and cultural mapping efforts.

137

Figure 42. Pouches of tobacco wrapped around a juniper tree, coincidentally discovered during field
sampling. This site, with a stunning view of Fountain Peak and the expanse of the Xáxli'p Survival
Territory, added an unexpected moment of reflection to the day’s work.

Development of the Web Mapping Tool
The web mapping platform was built using Esri ArcGIS Online and specifically
designed to meet the community's needs in wildfire management, land stewardship, and
cultural continuity. It is an important step in the ongoing journey toward accessible and
integrated community-directed land management. The development process was shaped by
the community-directed approach of this research partnership, incorporating continuous
feedback from Xáxli'p Community Forest crew members to ensure its relevance and
practicality. The fundamental guiding principle behind the tool was to make it as user-

138

friendly and adaptable as possible. The main interface provides a basic and intuitive layered
mapping view, allowing users to interact with various map layers and features, including
ALS-derived fuel maps, vegetation layers, landscape classifications, and cultural data (Figure
43). The interactive and customizable views help support activities such as wildfire risk
assessments, fire guard development, ecological planning, storytelling, and cultural
continuity.

Figure 43. Screenshot of the web mapping interface showing a zoomed-in view of the northeastern
section of the Xáxli'p Survival Territory with the canopy fuel layer and roads layer turned on. The
web map allows users to explore all fuel layers and spatial data in any web browser as long as they
have an internet connection. In this view, the canopy fuel layer highlights high fuel density areas in
red, medium density in orange/yellow, and lower density in green to dark green. The harvested areas
(cut blocks) in the northeastern portion of the study area stand out as green patches among higherdensity yellow, orange, and red areas, reflecting their lower canopy fuel load.

The collaborative design process that shaped this tool highlights its adaptability to the
evolving priorities of the Xáxli'p community. As their needs grow and change, the platform
can incorporate additional data types and functionalities, ensuring its adaptability and range
of uses for future iterations as technologies advance and community needs evolve. By
combining modern mapping capabilities with accessible design, this approach empowers the

139

Xáxli'p community to manage their lands with greater autonomy while acknowledging that it
represents one iteration in a broader effort to refine and enhance access to critical data and
resources.

Figure 44. A view of the Marble Canyon face from just north of the Xáxli'p Survival Territory. The
view was captured during the first field sampling campaign, with yellowing larch trees in the
foreground.

This democratization of information not only reduces reliance on external technical
support but also strengthens the community’s capacity for self-directed stewardship. As new
challenges emerge, the Xáxli'p are equipped with the skills and tools needed to respond
effectively, reinforcing their long-term commitment to autonomy and sustainable land
stewardship.
A particularly tangible outcome of this project was the hiring of new young XCF staff
from the community who took on responsibility for overseeing the technical aspects of the
web mapping tool, data collection workflow, and training the rest of the crew on how to use
the mapping interface developed through this work. This speaks directly to the issue of
cultural continuity, as it ensures that the knowledge, skills, and tools created through this
140

research remain within the community and continue to support Xáxli'p land stewardship for
generations to come. These hires represent one of the most concrete indicators of the project's
success in meeting its initial goal: strengthening local capacity, supporting self-determination
in land management, and creating something genuinely useful and lasting.
By creating a mapping platform that is both accessible and technically advanced, the
project has empowered community members to manage their land and wildfire risks with
greater autonomy and insight. Local forest crews and land stewards have developed an
enhanced ability to independently gather, interpret, and act on map layers and spatial
information directly related to the work they do in the field. This democratization of
information not only reduces reliance on external technical support but also strengthens the
community’s capacity for self-directed stewardship.
An Evolving Story
This research partnership began as an ambitious endeavor with seeming independent
objectives, including Indigenous knowledge and histories, community-directed research,
wildfire science, forest management, machine learning, and advanced remote sensing
technologies. At the outset, the work was expected to present technical challenges and was
approached as a cross-cultural partnership to promote co-learning about landscape-level
wildfire risk management. While it introduced numerous challenges and learning experiences
in programming, statistics, and analytical methodologies, it also evolved into a significant
chapter of growth, resilience, and transformation for both community and research partners.
An important note in understanding this evolving story is that this journey began
during an unprecedented global change. The arrival of COVID-19 disrupted not only global
systems but also the fundamental ways people connect, learn, and work. Classes shifted
141

online, fieldwork was delayed, and newly forming relationships were either paused or moved
to virtual spaces. Yet, despite these challenges, the research partnership remained a grounding
force. A brief window of opportunity allowed for the completion of fieldwork between
lockdowns, during which relationships with Xáxli’p community members were strengthened.
Working alongside the forest crew, elders, and knowledge holders who care deeply for their
survival territory and cultural continuity brought invaluable insights that shaped the project in
ways that could not have been anticipated.

Figure 45. View of the entrance to the Xáxli'p Community, looking up at Fountain Peak. This photo
was taken during the early days of the COVID-19 pandemic, with the Xáxli'p First Nation sign in the
foreground reading "Closed to the public, residents only."

142

Figure 46. Two female mule deer encountered during the first field sampling campaign in the Xáxli'p
Survival Territory.

As the world changed, so did the approach to this work. The technical aspects of the
project, designing a field sampling campaign, performing allometric calculations, learning to
code in R, and building machine learning models, were daunting. Entering the project with
only basic programming skills made the steep learning curve feel overwhelming, but it
reinforced the importance of patience, persistence, and iteration. Each challenge overcome
was a small victory, and each dataset analyzed brought a deeper understanding.
Yet the most significant lessons extended beyond the technical. This project was
never solely about creating maps; it was about stories. Maps serve as tools of cultural
communication, representations of reality that reflect the knowledge of the land and the
people who inhabit it. They are more than data, they are connections to the past, reflections
of the present, and guides for the future. Unearthing old maps and spatial data buried in filing
cabinets or lost on forgotten hard drives felt like recovering the voices of ancestors.
Revitalizing this information and making it accessible through a user-friendly, web-based

143

mapping platform for the Xáxli’p community became a mission and responsibility to make
the invisible visible and the inaccessible accessible.

Figure 47. A view of the Xáxli'p Community at the entrance of the Fountain Valley, with Fountain
Peak towering overhead. This photo was taken from the north side of the Fraser River Valley, looking
south into the Fountain Valley and the northwestern portion of the Xáxli'p Survival Territory.

The web mapping platform developed through this project became a bridge, linking
the life-sustaining knowledge of the past with the pressing needs of the present. It offered the
Xáxli’p forest crew new ways to engage with spatial data, integrating their local expertise
with broader spatial insights. Collaborative efforts explored how to bring maps into the field,
update them with real-time field-collected data, and apply them to ongoing fuel treatments,
invasive species monitoring, and the tracking of culturally significant features. These maps
became tools for connection and empowerment, designed to be accessible and interactive for
the historic inhabitants of Fountain Valley.
However, the web mapping tool is not a final outcome. It represents a valuable
resource and an important step forward. It is only one iteration in the ongoing effort to
enhance access to data and mapping for the Xáxli’p community. As technology advances,
these tools will continue to evolve, becoming more powerful and better aligned with the
144

community’s needs. While the current platform marks significant progress, it is not without
limitations. Though impactful, its present capabilities leave room for improvement. Future
iterations will undoubtedly expand its functionality and reach, offering engaging
opportunities for future partnerships.

Figure 48. Airborne view taken by drone over a sample plot during the initial field sampling
campaign. This photo, captured in the higher-elevation northeastern portion of the Xáxli'p Survival
Territory, shows a transition between three distinct forest types: dense lodgepole pine on the left, a
large, dense patch of spruce in the center, and a cut block densely replanted with young pine and
spruce.

Conclusions
My passion for technology and its potential to democratize knowledge has deepened
through this work. The emergence of generative AI, Large Language Models (LLMs), and
other interactive tools transformed how I approached challenges in coding, debugging,
analysis, and research. What once felt overwhelming became accessible and even exciting.
These technologies did not just help solve technical problems, they unlocked creativity,
accelerated learning, and sparked new possibilities.
Engaging with generative AI introduced a sense of freedom, making it easier to
articulate complex ideas and explore new intellectual directions. Instead of simply assisting

145

with tasks, these tools became creative collaborators, helping to shape raw thoughts into
structured, meaningful concepts. More than just a tool, AI has been a catalyst for deeper
thinking and innovation. This experience has reshaped my understanding of what is possible,
reinforcing my drive to explore how technology can bridge gaps, empower creativity, and
create meaningful change in both personal and global contexts.
As this project concludes, I reflect on what has been accomplished and how it has
shaped me. This work has taught me the value of persistence, the power of collaboration, and
the importance of centering community in research. It has shown me that maps are not just
tools but stories, and that these stories hold the power to connect people to their land, their
past, and their future.
My hope is that this research partnership has left a legacy for the Xáxli'p, a tool to
support their ongoing efforts toward land stewardship, cultural continuity, and wildfire
ecological resilience. I know that it has also left a legacy within me. The challenges I faced
and the lessons I learned are stepping stones on a path I am still discovering. The world is
profoundly different from when this journey began and will continue to evolve in ways we
cannot yet imagine. I am inspired to continue building tools and technologies that empower,
connect, and transform. This research partnership was an important part of my journey, and it
has solidified my belief in the power of maps, community, and technology to shape a better
future.
There are real-world implications about how spatial data, mapping technologies, and
AI-driven tools might be used in ways that do not align with the priorities of local
communities. This research helped illuminate the importance of keeping data in the hands of
the community, sharing knowledge on their terms, and developing tools that empower rather

146

than extract, ensuring scientific discovery and productivity remain connected to local
relevance and usefulness.
Reflecting on this project, I recognize that its significance extends beyond the maps
and models created. The real value lies in the relationships built, the broadened perspectives,
and the continued efforts to ensure that research serves the interests of those whose lands and
lives it seeks to support. The land is far more than a resource; it is a teacher, and learning
from it requires deep listening, patience, and respect. As I move forward, I carry with me the
understanding that the most impactful tools are built not for communities but with them.
Maps, like stories, hold the power to guide us all into an uncertain future with hope, and
purpose.

147

Figure 49. Researcher Patrick Robinson (me), loading truck with field gear after collecting a sample
plot during the field data collection campaign in the Xáxli'p Survival Territory. Thanks to the Xáxli'p
Community Forest team for providing the magnetic Xáxli'p Community Forest decal seen on the side
of Patrick’s truck used for access during fieldwork around the territory.

Final Acknowledgements
This research was conducted on the unceded traditional territory of the Xáxli'p
people, whose deep knowledge, stewardship, and connection to the land have shaped every
aspect of this work. I am profoundly grateful to the Xáxli'p Community Forest team and the
entire Xáxli'p community for their generosity, trust, and guidance throughout this project.
Their willingness to share knowledge, collaborate, and support this work has been
invaluable, and I am honored to have been part of this process.
I would also like to extend my sincere gratitude to my MSc supervisors, Scott Green
and Che Elkin, as well as my committee member, Marc Parisien, for their mentorship and
insight. Thank you to the UNBC GIS Lab, the Pacific Institute for Climate Solutions for
148

funding and support, and the team at Esri for their advice on mapping solutions. Lastly, I’m
deeply grateful to my friends and family, their encouragement and support helped carry me
through to the finish line.

To the Xáxli'p community, thank you for welcoming me, sharing your knowledge,
and allowing me to contribute in a small way to the important work you do.

149

References
Abatzoglou, J. T., & Williams, A. P. (2016). Impact of anthropogenic climate change on
wildfire across western US forests. Proceedings of the National Academy of Sciences,
113(42), 11770–11775.
Absolon, K., & Willett, C. (2004). Aboriginal Research: Berry Picking and Hunting in the
21st Century. First Peoples Child & Family Review, 1(1), Article 1.
Absolon, K., & Willett, C. (2005). Putting ourselves forward: Location in Aboriginal
research. In L. Brown & S. Strega (Eds.), Research as resistance: Critical,
Indigenous and anti-oppressive research approaches (pp. 97-125). Canadian
Scholars’ Press Inc.
Affleck, D. L. R., Keyes, C. R., & Goodburn, J. M. (2013). Conifer crown fuel modeling:
Current limits and potential for improvement. Western Journal of Applied Forestry,
28(4), 148–155.
Affleck, D., Seielstad, C., Goodburn, J., Queen, L., & Keane, R. (2013). Characterizing
crown biomass and crown profiles in conifer forests of the interior Northwest.
*Forest Ecology and Management*, 287, 1-12.
https://doi.org/10.1016/j.foreco.2012.08.014
Agee, J. K. & Skinner, C. N. (2005). Basic principles of forest fuel reduction treatments.
Forest Ecology and Management, 211(1-2), 83-96.
https://doi.org/10.1016/j.foreco.2005.01.034
Agee, J. K., & Skinner, C. N. (2005). Basic principles of forest fuel reduction treatments.
Forest Ecology and Management, 211(1–2), 83–96.
Ager, A., Vaillant, N., & Finney, M. (2010). A comparison of landscape fuel treatment
strategies to mitigate wildland fire risk in the urban interface and preserve old forest
structure. Forest Ecology and Management, 259(8), 1556-1570.
https://doi.org/10.1016/j.foreco.2010.01.032
Ahmed, O. S., Franklin, S. E., Wulder, M. A., & White, J. C. (2015). Characterizing standlevel forest canopy cover and height using Landsat time series, samples of airborne
LiDAR, and the Random Forest algorithm. ISPRS Journal of Photogrammetry and
Remote Sensing, 101, 89-101.
Ahmed, O. S., Robinson, A., Walsh, R. P. D., & Bravo, M. M. (2015). Assessment of forest
biomass and canopy structure using ALS in a subtropical forest. Remote Sensing of
Environment, 169, 170–180.

150

Alcasena, F., Salis, M., Ager, A. A., Castell, R., & Vega‐García, C. (2017). Assessing
wildland fire risk transmission to communities in northern spain. Forests, 8(2), 30.
https://doi.org/10.3390/f8020030
Alexander, M. E., & Cruz, M. G. (2013). Are the applications of wildland fire behaviour
models getting ahead of their evaluation again? Environmental Modelling &
Software, 41, 65–71.
Allen, I., Pawlikowski, N. C., Chhin, S., Premer, M., & Zhang, J. (2023). Modeling juvenile
stand development and fire risk of post-fire planted forests under variations in
thinning and fuel treatments using fvs–ffe. Forests, 14(6), 1223.
https://doi.org/10.3390/f14061223
Andersen, H.-E., McGaughey, R. J., & Reutebuch, S. E. (2005). Estimating forest canopy
fuel parameters using LIDAR data. Remote Sensing of Environment, 94(4), 441–449.
https://doi.org/10.1016/j.rse.2004.10.013
Anselin, L. (1995). Local Indicators of Spatial Association—LISA. Geographical Analysis,
27(2), 93–115.
Aragoneses, E. & Chuvieco, E. (2021). Generation and mapping of fuel types for fire risk
assessment. Fire, 4(3), 59. https://doi.org/10.3390/fire4030059
Arcidiacono, C., Procentese, F., & Di Napoli, I. (2014). Qualitative and quantitative research:
An ecological approach. International Journal of Multiple Research Approaches,
3(2), 163-176. https://doi.org/10.5172/mra.3.2.163
Baron, J. N., Hessburg, P. F., Parisien, M., Greene, G. A., Gergel, S. E., & Daniels, L. D.
(2024). Fuel types misrepresent forest structure and composition in interior British
Columbia: A way forward. Fire Ecology, 20(1). https://doi.org/10.1186/s42408-02400249-z
Baron, J. N., Hessburg, P. F., Parisien, M.-A., Greene, G. A., Gergel, S. E., & Daniels, L. D.
(2024). Fuel types misrepresent forest structure and composition in interior British
Columbia: A way forward. Fire Ecology, 20, 15.
Battaglia, M. A., & Shepperd, W. D. (2007). Ponderosa pine, mixed conifer, and spruce-fir
forests (S. M. Hood & M. Miller, Eds., p. 37). U.S. Department of Agriculture, Forest
Service, Rocky Mountain Research Station.
BC Ministry of Forests, L., Natural Resource Operations and Rural Development. (2020).
Lillooet timber supply area: Timber supply review data package. Government of
British Columbia.
Bechtold, W. A., & Patterson, P. L. (2005). The enhanced forest inventory and analysis
program—National sampling design and estimation procedures. U.S. Department of
Agriculture, Forest Service, Southern Research Station.
151

Benali, A., Sá, A. C. L., Pinho, J., Fernandes, P. M., & Pereira, J. M. C. (2021).
Understanding the impact of different landscape-level fuel management strategies on
wildfire hazard in central portugal. Forests, 12(5), 522.
https://doi.org/10.3390/f12050522
Berkes, F., Colding, J., & Folke, C. (2000). Rediscovery of traditional ecological knowledge
as adaptive management. Ecological Applications, 10(5), 1251–1262.
Beverly, J. L., & Forbes, A. M. (2023). Assessing directional vulnerability to wildfire.
Natural Hazards, 117(1), 831-849. https://doi.org/10.1007/s11069-023-05885-3
Bezzola, A. (2020). Incorporating an ethic of context and place in mechanistic research: A
place-based wildfire risk assessment in the Xáxli'p survival territory (Master's thesis,
University of Northern British Columbia).
Biau, G., & Scornet, E. (2016). A random forest-guided tour. Test, 25(2), 197–227.
Boots, B. (2003). Developing local measures of spatial association for categorical data.
Journal of Geographical Systems, 5(2), 139-160. https://doi.org/10.1007/s10109-0030110-3
Boulanger, Y., Gauthier, S., & Burton, P. J. (2014). A refinement of models projecting future
Canadian fire regimes using homogeneous fire regime zones. Canadian Journal of
Forest Research, 44(4), 365-376. https://doi.org/10.1139/cjfr-2013-0372
Bowman, D. M. J. S., Kolden, C. A., Abatzoglou, J. T., Johnston, F. H., van der Werf, G. R.,
& Flannigan, M. (2020). Vegetation fires in the Anthropocene. Nature Reviews Earth
& Environment, 1, 500–515.
Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5-32.
Bright, B. C., Hudak, A. T., Meddens, A. J. H., Hawbaker, T. J., Briggs, J. S., & Kennedy, R.
E. (2017). Prediction of forest canopy and surface fuels from ALS and satellite time
series data in a bark beetle-affected forest. Forests, 8(9), 322.
Bright, B. C., Hudak, A. T., Meddens, A. J. H., Hawbaker, T. J., Briggs, J. S., & Kennedy, R.
E. (2017). Prediction of forest canopy and surface fuels from LiDAR and satellite
time series data in a bark beetle-affected forest. Forests, 8(9), 322.
Cameron, H., Schroeder, D., & Beverly, J. L. (2021). Predicting black spruce fuel
characteristics with Airborne Laser Scanning (ALS). International Journal of
Wildland Fire, 31(2), 124-135. https://doi.org/10.1071/wf21004
Canadian Wildland Fire Information System. (n.d.). FBP SYSTEM fuel type descriptions.
Retrieved from https://cwfis.cfs.nrcan.gc.ca/background/fueltypes/c1

152

Castleden, H., Garvin, T., & Huu-ay-aht First Nation. (2012). Modifying Photovoice for
community-based participatory Indigenous research. Social Science & Medicine,
66(6), 1391–1402.
Charnley, S., Fischer, A. P., & Jones, E. T. (2007). Integrating traditional and local ecological
knowledge into forest biodiversity conservation in the Pacific Northwest. Forest
Ecology and Management, 246(1), 14-28.
https://doi.org/10.1016/j.foreco.2007.03.047
Charnley, S., Kelly, E. C., & Wendel, K. L. (2017). All lands approaches to fire management
in the Pacific West: A typology. Journal of Forestry, 115(1), 16–25.
Chen, Q., Vaglio Laurin, G., Valentini, R., & Yue, S. (2016). Aboveground biomass
estimation of Mediterranean forests with ALS and ground data. Remote Sensing, 8(1),
59.
Chen, Y., Zhu, X., Yebra, M., Harris, S., & Tapper, N. (2016). Strata-based forest fuel
classification for wildfire hazard assessment using terrestrial LiDAR. Journal of
Applied Remote Sensing, 10(4), 046025.
Christianson, A. (2015). Social science research on indigenous wildfire management in the
21st century and future research needs. International Journal of Wildland Fire, 24(2),
190–200. https://doi.org/10.1071/WF13048
Coogan, S. C. P., Robinne, F.-N., Jain, P., & Flannigan, M. D. (2019). Scientists’ warning on
wildfire—A Canadian perspective. Canadian Journal of Forest Research, 49(9),
1015–1023.
Coops, N. C., Hilker, T., Wulder, M. A., & St-Onge, B. (2016). Estimating canopy structure
of Douglas-fir forest stands from discrete-return ALS. Remote Sensing of
Environment, 98(4), 45–56.
Coops, N. C., Tompalski, P., Nijland, W., Rickbeil, G. J., Nielsen, S. E., Bater, C. W., &
Stadt, J. J. (2016). A forest structure habitat index based on airborne laser scanning
data. Ecological Indicators, 67, 346-357.
Copes‐Gerbitz, K., Dickson‐Hoyle, S., Ravensbergen, S. L., Hagerman, S., Daniels, L. D., &
Coutu, J. (2022). Community engagement with proactive wildfire management in
British Columbia, Canada: Perceptions, preferences, and barriers to action. Frontiers
in Forests and Global Change, 5. https://doi.org/10.3389/ffgc.2022.829125
Cowman, T. F., & Russell, A. (2021). Fuel load, stand structure, and understory species
composition following prescribed fire in an old-growth coast redwood (Sequoia
sempervirens) forest. *Fire Ecology*, 17(1), 1-15. https://doi.org/10.1186/s42408021-00098-0 2

153

Crotteau, J. S., Keyes, C. R., Hood, S. M., & Larson, A. J. (2019). Vegetation dynamics
following compound disturbance in a dry pine forest: Fuel treatment then bark beetle
outbreak. Ecological Applications, 30(2). https://doi.org/10.1002/eap.2023
Cruz, M. G., & Alexander, M. E. (2014). Canopy-fuel characteristics of conifer forests. Fire
Management Today, 73(4), 12-16.
Cruz, M. G., Alexander, M. E., & Wakimoto, R. H. (2003). Assessing canopy fuel stratum
characteristics in crown fire-prone fuel types of western North America. International
Journal of Wildland Fire, 12(1), 39-50. https://doi.org/10.1071/wf02024
Cutler, D. R., Edwards Jr, T. C., Beard, K. H., Cutler, A., Hess, K. T., Gibson, J., & Lawler, J.
J. (2007). Random forests for classification in ecology. Ecology, 88(11), 2783–2792.
Demarchi, D. A. (2011). The British Columbia Ecoregion Classification (3rd ed.). Ecosystem
Information Section, Ministry of Environment.
Diver, S. (2016). Community voices, the making and meaning of the Xáxli’p Community
Forest. A Report to the Xáxli’p Community Forest.
http://www.xcfc.ca/media/Community_Voices_Final_Report_2016_Full.pdf
Diver, S. (2016a). Community-based fire management in the Xaxli’p Community Forest.
Environmental Management, 57(1), 1–14. https://doi.org/10.1007/s00267-016-0771-3
Diver, S. (2016b). Co-management as a catalyst: Pathways to post-colonial forestry in the
Klamath Basin, California. Human Ecology, 44(5), 533–546.
https://doi.org/10.1007/s10745-016-9851-8
Diver, S. (2016c). Negotiating indigenous knowledge at the science-policy interface: Insights
from the Xáxli’p Community Forest. Environmental Science & Policy, 62, 34-44.
https://doi.org/10.1016/j.envsci.2016.03.001
Diver, S. (2016d). Community voices: The making and meaning of the Xáxli’p Community
Forest. University of California, Berkeley. https://www.xcfc.ca/community-voices/
Diver, S. (2017). Negotiating indigenous knowledge at the science-policy interface: Insights
from the Xáxli’p Community Forest. Environmental Science & Policy, 73, 1-11.
https://doi.org/10.1016/j.envsci.2017.03.001
Doerr, S. H., & Santín, C. (2016). Global trends in wildfire and its impacts: Perceptions
versus realities in a changing world. Philosophical Transactions of the Royal Society
B: Biological Sciences, 371(1696), 20150345.
Domingo, D., Riva, J. d. l., Gracia, M. T. L., García-Martín, A., Benlloch, P. I., Echeverría,
M., & Hoffrén, R. (2020). Fuel type classification using airborne laser scanning and
Sentinel 2 data in Mediterranean forests affected by wildfires. *Remote Sensing,
12*(21), 3660. https://doi.org/10.3390/rs12213660

154

Duff, T. J., Keane, R. E., Penman, T. D., & Tolhurst, K. G. (2017). Revisiting wildland fire
fuel quantification methods: The challenge of understanding a dynamic, biotic entity.
Forests, 8(9), 351.
Engelstad, P., Falkowski, M. J., Wolter, P. T., Poznanovic, A. J., & Johnson, P. (2019).
Estimating canopy fuel attributes from low-density lidar. Fire, 2(3), 38.
https://doi.org/10.3390/fire2030038
EpiCollect. (n.d.). EpiCollect: Data collection tool. Retrieved from https://five.epicollect.net/
Ermakov, A. I., & Kovalsky, S. V. (2018). Effect of clear-cut logging on forest stand
dynamics in northern Russia. 201, 012009. https://doi.org/10.1088/17551315/201/1/012009
Esri. (2021). ArcGIS Desktop: Release 10.8.1. Environmental Systems Research Institute.
Fernández-Álvarez, M., Armesto, J., & Picos, J. (2019). Lidar-based wildfire prevention in
WUI: The automatic detection, measurement and evaluation of forest fuels. Forests,
10(2), 148. https://doi.org/10.3390/f10020148
Fernández-Guisuraga, J. M., González, J. A., & García, M. (2020). LiDAR-based wildfire
prevention in WUI: The automatic detection, measurement and evaluation of forest
fuels. *Forests*, 11(22), 2638. https://doi.org/10.3390/f11222638
Finney, M. A. (2005). The challenge of quantitative risk analysis for wildland fire. Forest
Ecology and Management, 211(1–2), 97–108.
Finney, M. A., McHugh, C. W., Grenfell, I. C., Riley, K. L., & Short, K. C. (2011). A
simulation of probabilistic wildfire risk components for the continental United States.
Stochastic Environmental Research and Risk Assessment, 25(7), 973–1000.
First Nations Information Governance Centre. (2014). Ownership, control, access, and
possession (OCAPTM): The path to First Nations information governance.
https://fnigc.ca/ocap-training/
Flannigan, M. D., Krawchuk, M. A., De Groot, W. J., Wotton, B. M., & Gowman, L. M.
(2009). Implications of changing climate for global wildland fire. International
Journal of Wildland Fire, 18(5), 483-507.
Forbes, B., Reilly, S., Clark, M. L., Ferrell, R., Kelly, A., Krause, P., … & Bentley, L. P.
(2022). Comparing remote sensing and field-based approaches to estimate ladder
fuels and predict wildfire burn severity. Frontiers in Forests and Global Change, 5.
https://doi.org/10.3389/ffgc.2022.818713
Forest Analysis and Inventory Branch. (2022). Lillooet timber supply area: Timber supply
analysis discussion papeR. Ministry of Forests.
https://www2.gov.bc.ca/assets/gov/environment/plants-animals-and-

155

ecosystems/ecosystems/broadecosystem/an_introduction_to_the_ecoregions_of_british_columbia.pdf
Forest Analysis and Inventory Branch. (2022). Provincial forest inventory. BC Ministry of
Forests.
Forestry Canada Fire Danger Group. (1992). Development and structure of the Canadian
Forest Fire Behavior Prediction System. Information Report ST-X-3. Forestry
Canada, Science, and Sustainable Development Directorate.
FPInnovations. (2020). Tools and techniques to forest fuel. Retrieved from
https://library.fpinnovations.ca/media/FOP/TR2020N43.PDF
Gajardo, J., García, M., & Riaño, D. (2014). Applications of Airborne Laser Scanning in
forest fuel assessment and fire prevention. In Forestry applications of Airborne Laser
Scanning (pp. 439-462). Springer, Dordrecht.
García-Cimarras, A., Manzanera, J. A., & Valbuena, R. (2021). Analysis of Mediterranean
vegetation fuel type changes using multitemporal lidar. Forests, 12(3), 335.
https://doi.org/10.3390/f12030335
Gayton, D. V. (2015). Documenting fire history in a British Columbia Ecological Reserve.
Journal of Ecosystems & Management, 14(1).
GitHub. (n.d.). Copilot [AI coding assistant]. Retrieved from
https://github.com/features/copilot
Goetze, T. C. (2005). Empowered co-management: Towards power-sharing and indigenous
rights in Clayoquot Sound, BC. Anthropologica, 47, 247-265.
Golodets, C., Kigel, J., Sapir, Y., & Sternberg, M. (2012). Quantitative vs qualitative
vegetation sampling methods: A lesson from a grazing experiment in a Mediterranean
grassland. Applied Vegetation Science, 16(1), 147-157.
https://doi.org/10.1111/avsc.12005
González‐Ferreiro, E., Arellano-Pérez, S., Castedo‐Dorado, F., Hevia, A., Vega, J. A., VegaNieva, D. J., … & Ruiz-González, A. D. (2017). Modelling the vertical distribution of
canopy fuel load using national forest inventory and low-density airborne laser
scanning data. Plos One, 12(4), e0176114.
https://doi.org/10.1371/journal.pone.0176114
Google. (n.d.). AI tools. Retrieved from https://ai.google/
Government of British Columbia. (2021). Fuel management prescription guidance.
Retrieved from https://www2.gov.bc.ca/assets/gov/public-safety-and-emergencyservices/wildfire-status/prevention/fire-fuel-management/fuelsmanagement/2021_fuel_management_prescription_guidance.pdf

156

Government of Canada. (2023). Climate normals (1991-2020) for British Columbia.
Meteorological Service of Canada.
https://climate.weather.gc.ca/climatenormals/results19912020e.html?searchType=stn
Prov&lstProvince=BC&txtCentralLatMin=0&txtCentralLatSec=0&txtCentralLongM
in=0&txtCentralLongSec=0&stnID=336000000&dispBack=0
Grammarly. (n.d.). Generative AI. Retrieved from https://www.grammarly.com/
Groot, W. J. D., Hanes, C. C., & Wang, Y. (2022). Crown fuel consumption in Canadian
boreal forest fires. International Journal of Wildland Fire, 31(3), 255-276.
https://doi.org/10.1071/wf21049
Hanes, C. C., Wang, X., & Groot, W. J. D. (2021). Dead and down woody debris fuel loads in
Canadian forests. International Journal of Wildland Fire, 30(11), 871-885.
https://doi.org/10.1071/wf21023
Hastie, T., Tibshirani, R., & Friedman, J. (2009). The elements of statistical learning: Data
mining, inference, and prediction (2nd ed.). Springer Series in Statistics. Springer.
He, L., Jeanneau, A., Ramsey, S., Radford, D. A. G., Zecchin, A. C., Reinke, K., Jones, S. D.,
van Delden, H., McNaught, T., Westra, S., & Maier, H. R. (2024). Estimating fuel
load for wildfire risk assessment at regional scales using earth observation data: A
case study in Southwestern Australia. Remote Sensing Applications: Society and
Environment, 36, Article 101356.
Hessburg, P. F., Agee, J. K., & Franklin, J. F. (2005). Dry forests and wildland fires of the
inland Northwest USA: Contrasting the landscape ecology of the pre-settlement and
modern eras. Forest Ecology and Management, 211(1–2), 117–139.
https://doi.org/10.1016/j.foreco.2005.02.010
Hiers, J. K., O’Brien, J. J., Mitchell, R., Grego, J. M., & Loudermilk, E. L. (2009). The
wildland fuel cell concept: An approach to characterize fine-scale variation in fuels
and fire in frequently burned longleaf pine forests. International Journal of Wildland
Fire, 18(3), 315. https://doi.org/10.1071/wf08084
Hilker, T., Leeuwen, M. v., Coops, N. C., Wulder, M. A., Newnham, G., Jupp, D. L., … &
Culvenor, D. (2010). Comparing canopy metrics derived from terrestrial and airborne
laser scanning in a Douglas-fir dominated forest stand. Trees, 24(5), 819-832.
https://doi.org/10.1007/s00468-010-0452-7
Hoffman, K. M., Daniels, L. D., & Da Silva, L. (2021). Fire history in the dry forests of the
southern interior, British Columbia, Canada. Forest Ecology and Management, 479,
118595.
Hollaus, M., Wagner, W., & Karel, W. (2019). Airborne laser scanning for forest fire risk
assessment: A review of recent advances. *Remote Sensing of Environment*, 261,
112487. https://doi.org/10.1016/j.rse.2021.112487
157

Hopkinson, C., Chasmer, L., Young-Pow, C., & Treitz, P. (2004). Assessing forest metrics
with a ground-based scanning ALS. Canadian Journal of Forest Research, 34(3),
573–583.
Indigenous and Northern Affairs Canada. (2019, July). Xaxli’p registered population.
Retrieved August 14, 2019, from First Nations Profiles. http://fnpppn.aandcaadnc.gc.ca/fnp/Main/Search/FNRegPopulation.aspx?BAND_NUMBER=5
92&lang= eng
Indigenous and Northern Affairs Canada. (n.d.). Xaxli'p First Nation population profile.
Retrieved July 28, 2024, from https://fnp-ppn.aadncaandc.gc.ca/FNP/Main/Search/FNRegPopulation.aspx?BAND_NUMBER=592&lang
=eng
Ivey, M. A., Wonkka, C. L., Weidig, N. C., & Donovan, V. M. (2024). Woody cover fuels
large wildfire risk in the eastern us. Geophysical Research Letters, 51(24).
https://doi.org/10.1029/2024gl110586
Jakubowski, M. K., Guo, Q., & Kelly, M. (2013). Tradeoffs in ALS-based biomass
estimation across forest gradients in the Sierra Nevada, California. Remote Sensing of
Environment, 139, 56–68.
Jenkins, J. C., Chojnacky, D. C., Heath, L. S., & Birdsey, R. A. (2004). Comprehensive
database of diameter-based biomass regressions for North American tree species.
U.S. Department of Agriculture, Forest Service, Northeastern Research Station.
Johnson, J. T., Howitt, R., Cajete, G., Berkes, F., Louis, R. P., & Kliskey, A. (2015). Weaving
indigenous and sustainability sciences to diversify our methods. Sustainability
Science, 11(1), 1–11.
Jolly, W. M., Cochrane, M. A., Freeborn, P. H., Holden, Z. A., Brown, T. J., Williamson, G.
J., & Bowman, D. M. J. S. (2015). Climate-induced variations in global wildfire
danger from 1979 to 2013. Nature Communications, 6, 7537.
Kalabokidis, K., Palaiologou, P., Athanasis, N., Vasilakos, C., & Matsinos, Y. (2012). GIS
and remote sensing techniques for the assessment of forest fire risk in the
Mediterranean. Forests, 3(3), 666–685.
Keane, R. E. (2015). Wildland fuel fundamentals and applications. Springer.
Keane, R. E., Burgan, R., & van Wagtendonk, J. (2001). Mapping wildland fuels for fire
management across multiple scales: Integrating remote sensing, GIS, and biophysical
modeling. International Journal of Wildland Fire, 10(4), 301–319.
Keane, R. E., Cary, G. J., Davies, I. D., Flannigan, M. D., Gardner, R. H., Lavorel, S.,
Lenihan, J. M., Li, C., & Rupp, T. S. (2001). A classification of landscape fire
succession models: Spatial simulations of fire and vegetation dynamics. Ecological
Modelling, 144(1), 3–18. https://doi.org/10.1016/S0304-3800(01)00399-7
158

Kelly, M., Su, Y., Tommaso, S. D., Fry, D., Collins, B. M., Stephens, S. L., … & Guo, Q.
(2017). Impact of error in lidar-derived canopy height and canopy base height on
modeled wildfire behavior in the sierra nevada, california, usa. Remote Sensing,
10(1), 10. https://doi.org/10.3390/rs10010010
Kindling, M., & Strecker, D. (2022). Data quality assurance at research data repositories.
Data Science Journal, 21. https://doi.org/10.5334/dsj-2022-018
Klenner, W., & Arsenault, R. (2008). Dry forests in the Southern Interior of British
Columbia: Historic disturbances and implications for restoration and management.
*Forest Ecology and Management*, 256(3), 1–12.
https://doi.org/10.1016/j.foreco.2008.02.047
Kreye, J. K., Varner, J. M., & Dugaw, C. J. (2014). Spatial and temporal variability of forest
floor duff characteristics in long-unburned pinus palustris forests. Canadian Journal
of Forest Research, 44(12), 1477-1486. https://doi.org/10.1139/cjfr-2014-0223
Laushman, K. M., Munson, S. M., & Villarreal, M. L. (2020). Wildfire risk and hazardous
fuel reduction treatments along the US-Mexico border: A review of the science (19862019). Air, Soil and Water Research, 13. https://doi.org/10.1177/1178622120950272
Leydsman McGinty, E. (2024). Leveraging the synergistic power of Landsat and GEDI data
to support sustainable forest management. Landsat Science.
https://landsat.gsfc.nasa.gov/article/leveraging-synergistic-power-landsat-gedi-datasupport-sustainable-forest-management
Liang, X., Hyyppä, J., Kaartinen, H., Lehtomäki, M., Pyörälä, J., Pfeifer, N., Holopainen, M.,
Brolly, G., Francesco, P., Hackenberg, J., Huang, H., Jo, H.-W., Katoh, M., Liu, L.,
Mokroš, M., Morel, J., Olofsson, K., Poveda-Lopez, J., Trochta, J., & Wang, Y.
(2018). International benchmarking of terrestrial laser scanning approaches for forest
inventories. ISPRS Journal of Photogrammetry and Remote Sensing, 144, 137–179.
Liaw, A., & Wiener, M. (2002). Classification and regression by Random Forest. R News,
2(3), 18–22.
Liu, J., Wang, Y., Guo, H., et al. (2024). Spatial and temporal patterns and driving factors of
forest fires based on an optimal parameter-based geographic detector in the Panxi
region, Southwest China. Fire Ecology, 20, 27. https://doi.org/10.1186/s42408-02400257-z
Mallinis, G., Mitsopoulos, Ι., Beltran, E., & Goldammer, J. G. (2016). Assessing wildfire risk
in cultural heritage properties using high spatial and temporal resolution satellite
imagery and spatially explicit fire simulations: The case of Holy Mount Athos,
Greece. Forests, 7(2), 46. https://doi.org/10.3390/f7020046
Marcoux, K., Coogan, S. C., & Danaher, T. (2021). A disrupted historical fire regime in
Central British Columbia. *Forest Ecology and Management*, 490, 119–134.
https://doi.org/10.1016/j.foreco.2021.119134
159

Martin-Ducup, O., Dupuy, J. L., Soma, M., Guerra-Hernandez, J., Marino, E., Fernandes, P.
M., ... & Pimont, F. (2025). Unlocking the potential of Airborne LiDAR for direct
assessment of fuel bulk density and load distributions for wildfire hazard
mapping. Agricultural and Forest Meteorology, 362, 110341.
Masinda, M. M., Li, F., Liu, Q., Sun, L., & Hu, T. (2021). Forest fire risk estimation in a
typical temperate forest in northeastern China using the Canadian forest fire weather
index: Case study in autumn 2019 and 2020. Natural Hazards, 111(1), 1085-1101.
https://doi.org/10.1007/s11069-021-05054-4
Meidinger, D., & Pojar, J. (1991). Ecosystems of British Columbia. B.C. Ministry of Forests,
Victoria, B.C. Special Report Series 6.
Miller, C., & Thode, A. E. (2007). Quantifying burn severity in a heterogeneous landscape
with a relative version of the delta Normalized Burn Ratio (dNBR). Remote Sensing
of Environment, 109(1), 66-80.
Minasny, B., & McBratney, A. B. (2006). A conditioned Latin hypercube method for
sampling in the presence of ancillary information. Computers & Geosciences, 32(9),
1378–1388.
Ministry of Environment and Climate Change Strategy - Knowledge Management. (2019).
Ecosections—Ecoregion ecosystem classification of British Columbia [Data
Warehouse]. Retrieved June 19, 2019, from BC Data Catalogue website:
https://catalogue.data.gov.bc.ca/dataset/ecosections-ecoregionecosystemclassification-of-british-columbia
Montoro Girona, M., Fenton, N. J., Simard, M., & Bergeron, Y. (2017). Innovative
silviculture to achieve sustainable forest management in boreal forests: Lessons from
two large-scale experiments. Springer. https://link.springer.com/chapter/10.1007/9783-319-98313-7_9
Moran, P. (1950). A test for the serial independence of residuals. Biometrika, 37(1-2), 178181. http://dx.doi.org/10.1093/biomet/37.1-2.178
Moritz, M. A., Batllori, E., Bradstock, R. A., Gill, A. M., Handmer, J., Hessburg, P. F.,
Leonard, J., McCaffrey, S., Odion, D. C., Schoennagel, T., & Syphard, A. D. (2014).
Learning to coexist with wildfire. Nature, 515(7525), 58–66.
Mutlu, M., Popescu, S. C., & Zhao, K. (2008). Aboveground biomass estimation using ALS
in a temperate mixed forest in central Ontario, Canada. Remote Sensing of
Environment, 112(5), 2369–2380.
Mutlu, M., Popescu, S. C., Stripling, C., & Spencer, T. (2008). Mapping surface fuel models
using lidar and multispectral data fusion for fire behavior. Remote Sensing of
Environment, 112(1), 274-285.

160

Nadeau, L. B., McRae, D. J., & Jin, J. Z. (2005). Development of a national fuel-type map for
Canada using fuzzy logic. Information Report NOR-X-406. Natural Resources
Canada, Canadian Forest Service, Northern Forestry Centre.
Odion, D. C., Hanson, C. T., Arsenault, A., Baker, W. L., DellaSala, D. A., Hutto, R. L., … &
Williams, M. A. (2014). Examining historical and current mixed-severity fire regimes
in ponderosa pine and mixed-conifer forests of western North America. PLoS ONE,
9(2), e87852. https://doi.org/10.1371/journal.pone.0087852
OpenAI. (2022). *ChatGPT* Computer software. Available at:
https://www.openai.com/chatgpt
Ottmar, R. D., Sandberg, D. V., Riccardi, C. L., & Prichard, S. J. (2007). An overview of the
Fuel Characteristic Classification System—Quantifying, classifying, and creating fuel
beds for resource planning. Canadian Journal of Forest Research, 37(12), 2383-2393.
Parisien, M. A., Barber, Q. E., Bourbonnais, M. L., Daniels, L. D., Flannigan, M. D., Gray,
R. W., ... & Whitman, E. (2023). Abrupt, climate-induced increase in wildfires in
British Columbia since the mid-2000s. Communications Earth & Environment, 4(1),
309.
Parsons, R. A., Linn, R., Pimont, F., Hoffman, C., Sauer, J. A., Winterkamp, J., … & Jolly, W.
M. (2017). Numerical investigation of aggregated fuel spatial pattern impacts on fire
behavior. Land, 6(2), 43. https://doi.org/10.3390/land6020043
Pausas, J. G., & Keeley, J. E. (2019). Wildfires as an ecosystem service. Frontiers in Ecology
and the Environment, 17(5), 289–295.
Paveglio, T. B., Prato, T., Edgeley, C. M., & Nalle, D. J. (2016). Evaluating the
characteristics of social vulnerability to wildfire: Demographics, perceptions, and
parcel characteristics. Environmental Management, 58(3), 534-548.
https://doi.org/10.1007/s00267-016-0719-x
Perrakis, D. D. B., & Eade, G. (2015). British Columbia Wildfire Fuel Typing and Fuel Type
Layer Description. BC Wildfire Service, Ministry of Forests, Lands, and Natural
Resource Operations.
Phelps, N., & Beverly, J. (2024). Fuel types misrepresent forest structure and composition in
interior British Columbia, Canada. Fire Ecology, 20(1), 1-20.
Phelps, N., & Beverly, J. L. (2022). Classification of forest fuels in selected fire-prone
ecosystems of Alberta, Canada—implications for crown fire behaviour prediction and
fuel management. Annals of Forest Science, 79(1), 40.
Pichler, M., Boreux, V., Klein, A. M., Schleuning, M., & Hartig, F. (2023). Machine learning
and deep learning—A review for ecologists. Methods in Ecology and Evolution,
14(2), 475-494
161

Pierce, A. D., Farris, C. A., & Taylor, A. H. (2012). Use of random forests for modeling and
mapping forest canopy fuels for fire behavior analysis in Lassen Volcanic National
Park, California, USA. Forest Ecology and Management, 279, 77-89.
https://doi.org/10.1016/j.foreco.2012.05.010
Pimont, F., Parsons, R., Rigolot, E., de Coligny, F., Dupuy, J.-L., Dreyfus, P., & Linn, R. R.
(2016). Modeling fuels and fire effects in 3D: Model description and applications.
Environmental Modelling & Software, 80, 225-244.
https://doi.org/10.1016/j.envsoft.2016.03.003
Pojar, J., Klinka, K., & Meidinger, D. V. (1987). Biogeoclimatic ecosystem classification in
British Columbia. Forest Ecology and Management, 22(1), 119–154.
https://doi.org/10.1016/0378-1127(87)90100-9
Prentiss, A. M., & Kuijt, I. (2012). People of the Middle Fraser Canyon: An archaeological
history. UBC Press.
Probst, P., Wright, M. N., & Boulesteix, A.-L. (2019). Hyperparameters and tuning strategies
for random forest. Wiley Interdisciplinary Reviews: Data Mining and Knowledge
Discovery, 9(3), e1301.
Puettmann, K. J., Coates, K. D., & Messier, C. C. (2009). A critique of silviculture:
Managing for complexity. Island Press.
R Core Team. (2020). R: A language and environment for statistical computing. R
Foundation for Statistical Computing.
Raymond, C. L., & Peterson, D. L. (2005). Fuel treatments alter the effects of wildfire in a
mixed-evergreen forest, Oregon, USA. Canadian Journal of Forest Research, 35(12),
2981-2995. https://doi.org/10.1139/x05-206
Reilly, S., Clark, M. L., Bentley, L. P., Matley, C., Piazza, E., & Oliveras, I. (2021). The
potential of multispectral imagery and 3d point clouds from unoccupied airborne
systems (UAS) for monitoring forest structure and the impacts of wildfire in
Mediterranean-climate forests. Remote Sensing, 13(19), 3810.
https://doi.org/10.3390/rs13193810
Reinhardt, E. D. (2006). Using FOFEM and FVS to estimate forest floor and woody fuel
consumption. Fire Management Today, 66(1), 45–50.
Reinhardt, E., Scott, J., Gray, K., & Keane, R. (2006). Estimating canopy fuel characteristics
in five conifer stands in the western United States using tree and stand
measurements. Canadian Journal of Forest Research, 36(11), 2803-2814.
Reutebuch, S. E., McGaughey, R. J., Andersen, H.-E., & Carson, W. W. (2003). Accuracy of
a high-resolution ALS terrain model under a conifer forest canopy. Canadian Journal
of Remote Sensing, 29(5), 527–535.
162

Riegl Laser Measurement Systems GmbH. (2018). RIEGL VQ-580 II: Waveform Processing
Airborne Laser Scanning System. Retrieved from
http://www.riegl.com/nc/products/airbornescanning/produktdetail/product/scanner/65/
Riegl. (2020). RIEGL VQ-580 II Airborne Laser Scanners.
https://products.rieglusa.com/item/airborne-scanners/riegl-vq-580-ii-airborne-laserscanners
Rothermel, R. C. (1983). How to predict the spread and intensity of forest and range fires.
U.S. Department of Agriculture, Forest Service, Intermountain Forest and Range
Experiment Station.
Roudier, P., Beaudette, D. E., Hewitt, A. E., & McBratney, A. B. (2012). A conditioned Latin
hypercube sampling algorithm incorporating operational constraints. Geoderma, 181,
45–52.
Roudier, P., Gascuel-Odoux, C., & Durrieu, S. (2014). A conditioned Latin hypercube
sampling algorithm incorporating operational constraints. Environmental Modelling
& Software, 62, 1-10. https://doi.org/10.1016/j.envsoft.2014.08.001
Roussel, J. R., & Auty, D. (2020). LidR: Airborne ALS Data Manipulation and Visualization
for Forestry Applications.
Roussel, J. R., Auty, D., Coops, N. C., Tompalski, P., Goodbody, T. R. H., Sánchez Meador,
A., Bourdon, J.-F., de Boissieu, F., & Achim, A. (2020). lidR: An R package for
analysis of Airborne Laser Scanning (ALS) data. Remote Sensing of Environment,
251, 112061. https://doi.org/10.1016/j.rse.2020.112061
Schnarch, B. (2004). Ownership, control, access, and possession (OCAP) or selfdetermination applied to research: A critical analysis of contemporary First Nations
research and some options for First Nations communities. Journal of Aboriginal
Health, 1(1), 80-95.
Schoennagel, T., Veblen, T. T., Negrón, J. F., & Smith, J. M. B. (2012). Effects of mountain
pine beetle on fuels and expected fire behavior in lodgepole pine forests, Colorado,
USA. PLoS ONE, 7(1), e30002. https://doi.org/10.1371/journal.pone.0030002
Scott, J. H., & Reinhardt, E. D. (2001). Assessing crown fire potential by linking models of
surface and crown fire behavior. USDA Forest Service Rocky Mountain Research
Station Research Paper RMRS-RP-29.
Shirk, J., Ballard, H., Wilderman, C., Phillips, T., Wiggins, A., Jordan, R., McCallie, E.,
Minarchek, M., Lewenstein, B., Krasny, M., & Bonney, R. (2012). Public
participation in scientific research: A framework for deliberate design. Ecology and
Society, 17(2), 29-48. https://doi.org/10.5751/ES-04705-170229

163

Skowronski, N. S., Clark, K. L., Nelson, R. F., Hom, J. L., & Patterson, M. (2015). Structurelevel fuel load assessment in the wildland-urban interface: A comparison of
methodologies and operational implications. International Journal of Wildland Fire,
25(5), 547–557.
Spits, C., Wallace, L., & Reinke, K. (2017). Investigating surface and near-surface bushfire
fuel attributes: A comparison between visual assessments and image-based point
clouds. Sensors, 17(4), 910. https://doi.org/10.3390/s17040910
Spoon, J. (2014). Quantitative, qualitative, and collaborative methods: Approaching
indigenous ecological knowledge heterogeneity. Ecology and Society, 19(3), Article
33. https://www.jstor.org/stable/26269613
Stephens, S. L., Moghaddas, J. J., Edminster, C. B., Fiedler, C. E., Haase, S. M., Harrington,
M. G., … & Youngblood, A. (2009). Fire treatment effects on vegetation structure,
fuels, and potential fire severity in western U.S. forests. Ecological Applications,
19(2), 305-320. https://doi.org/10.1890/07-1755.1
Stocks, B. J., Mason, J. A., Todd, J. B., Bosch, E. M., Wotton, B. M., Amiro, B. D., & Logan,
K. A. (1989). Fire behavior in mature jack pine. Canadian Journal of Forest
Research, 19(6), 783–790.
Stocks, B. J., Wotton, B. M., & McRae, D. J. (1989). Canadian Forest Fire Behavior
Prediction System: User's Guide. Canadian Forestry Service.
Stupariu, M. S., Cushman, S. A., Plesoianu, A. I., Patru-Stupariu, I., & Furst, C. (2022).
Machine learning in landscape ecological analysis: A review of recent approaches.
Landscape Ecology, 37, 1227-1250
Syphard, A. D., & Keeley, J. E. (2020). Why are so many structures burning in California?
Fremontia, 48(2), 28–35.
TallBear, K. (2013). Native American DNA: Tribal belonging and the false promise of genetic
science. University of Minnesota Press.
Taylor, S. W., & Alexander, M. E. (2006). Science, technology, and human factors in fire
danger rating: The Canadian experience. International Journal of Wildland Fire,
15(2), 121–135.
The First Nations Information Governance Centre. (n.d.). Home. Retrieved July 28, 2024,
from https://fnigc.ca/
Turner, N. J., Ignace, M. B., & Ignace, R. (2000). Traditional ecological knowledge and
wisdom of Aboriginal peoples in British Columbia. Ecological Applications, 10(5),
1275–1287.
Ubcwiki. (2021). Comparing the regulations and management practices of community
forestry tenures in BC: The Canim Lake First Nations Woodland License, the Xaxli’p
164

Community Forest Agreement, and the South Canoe Woodlot. Retrieved July 28,
2024, from
https://wiki.ubc.ca/Course:FRST370/2021/Comparing_the_regulations_and_manage
ment_practises_of_community_forestry_tenures_in_BC:_The_Canim_Lake_First_Na
tions_Woodland_License,_the_Xaxli%E2%80%99p_Community_Forest_Agreement,
_and_the_South_Canoe_Woodlot#cite_note-:3-15
Ung, C.-H., Bernier, P., & Guo, X.-J. (2008). Canadian national biomass equations: New
parameter estimates that include British Columbia Data. Canadian Journal of Forest
Research, 38(5), 1123–1132. https://doi.org/10.1139/x07-224
Ung, C.-H., Bernier, P., Guo, X. J., & Lambert, M. C. (2008). Allometric equations of
biomass for 22 tree species in eastern Canada. Annals of Forest Science, 65(4), 1–10.
Vafaei, S., Soosani, J., Adeli, K., Fadaei, H., Naghavi, H., Pham, T. D., … & Bui, D. T.
(2018). Improving accuracy estimation of forest aboveground biomass based on
incorporation of alos-2 palsar-2 and sentinel-2a imagery and machine learning: A case
study of the Hyrcanian forest area (Iran). Remote Sensing, 10(2), 172.
https://doi.org/10.3390/rs10020172
van Ewijk, K. Y., Sluiter, R., Barendregt, A., & Schmidt, A. M. (2011). Tree biomass
estimation using airborne ALS in a tropical forest: A comparison study. Remote
Sensing of Environment, 115(10), 2354–2361.
van Ewijk, K. Y., Treitz, P. M., & Scott, N. A. (2011). Characterizing forest succession in
Central Ontario using lidar-derived indices. Photogrammetric Engineering & Remote
Sensing, 77(3), 261-269
Van Wagner, C. E. (1987). Development and structure of the Canadian Forest Fire Weather
Index System. Canadian Forest Service.
Vastaranta, M., Kankare, V., Holopainen, M., Yu, X., Hyyppä, J., & Hyyppä, H. (2012).
Combination of individual tree detection and area-based approach in imputation of
forest variables using airborne laser data. ISPRS Journal of Photogrammetry and
Remote Sensing, 67, 73-79
Vastaranta, M., Kankare, V., Wulder, M. A., White, J. C., & Holopainen, M. (2012). Areabased approach for forest inventory using airborne laser scanning and multispectral
imagery. Forests, 3(3), 570–590.
Viedma, O., Silva, C. A., Moreno, J. M., & Hudak, A. T. (2024). Ladder fuels: A new
automated tool for vertical fuel continuity analysis and crown base height detection
using light detection and ranging. Methods in Ecology and Evolution, 15(11), 19581967. https://doi.org/10.1111/2041-210x.14427
Wang, S. and Niu, S. (2016). Fuel classes in conifer forests of southwest Sichuan, China, and
their implications for fire susceptibility. Forests, 7(3), 52.
https://doi.org/10.3390/f7030052
165

Wasserman, T. N., & Mueller, S. E. (2023). Climate influences on future fire severity: a
synthesis of climate-fire interactions and impacts on fire regimes, high-severity fire,
and forests in the western United States. Fire Ecology, 19, 43.
https://doi.org/10.1186/s42408-023-00200-8
Weinstein, M. (1995). Thinking like an inhabitant: Understanding Xáxli’p environmental
values. Prepared for Xáxli’p Nation—BC Government Joint Land Use Planning
Workshop.
Westerling, A. L. (2016). Increasing western US forest wildfire activity: Sensitivity to
changes in the timing of spring. Philosophical Transactions of the Royal Society B:
Biological Sciences, 371(1696), 20150178.
White, J. C., Coops, N. C., Wulder, M. A., Vastaranta, M., Hilker, T., & Tompalski, P. (2016).
Remote sensing technologies for enhancing forest inventories: A review. Canadian
Journal of Remote Sensing, 42(5), 619-641.
https://doi.org/10.1080/07038992.2016.1207484
White, J. C., Wulder, M. A., Varhola, A., Vastaranta, M., Coops, N. C., Cook, B. D., Pitt, D.,
& Woods, M. (2013a). A best practices guide for generating forest inventory attributes
from airborne laser scanning data using an area-based approach. The Forestry
Chronicle, 89(6), 722–733.
White, J. C., Wulder, M. A., Varhola, A., Vastaranta, M., Coops, N. C., Cook, B. D., Pitt, D.,
& Woods, M. (2013b). A best practices guide for generating forest inventory
attributes from airborne laser scanning data using an area-based approach (Version
2.0). Canadian Forest Service, Canadian Wood Fibre Centre.
White, J. C., Wulder, M. A., Vastaranta, M., Coops, N. C., Pitt, D., & Woods, M. (2013). The
utility of image-based point clouds for forest inventory: A comparison with airborne
laser scanning. Forests, 4(3), 518–536. https://doi.org/10.3390/f4030518
Woods, L. K., & Treitz Paul. (2008). Predicting forest stand variables from ALS data in the
Great Lakes – St. Lawrence forest of Ontario. The Forestry Chronicle, 84(6), 827–
839. https://doi.org/10.5558/tfc84827-6
Woods, M., Pitt, D., Penner, M., Lim, K., Nesbitt, D., Etheridge, D., & Treitz, P. (2011).
Operational implementation of a LiDAR inventory in Boreal Ontario. The Forestry
Chronicle, 87(4), 512-528.
Wotton, B. M. (2008). Interpreting and using outputs from the Canadian forest fire danger
rating system in research applications. Environmental and Ecological Statistics,
16(2), 107-131. https://doi.org/10.1007/s10651-007-0084-2
Wotton, B. M., Martell, D. L., & Logan, K. A. (2010). Climate change and people-caused
forest fire occurrence in Ontario. Climatic Change, 104, 55–74.

166

Wulder, M. A., White, J. C., Loveland, T. R., Woodcock, C. E., Belward, A. S., Cohen, W. B.,
& et al. (2016). The global Landsat archive: Status, consolidation, and direction.
*Remote Sensing of Environment*, 185, 271–283.
https://doi.org/10.1016/j.rse.2016.03.014
Wulder, M. A., White, J. C., Nelson, R. F., Næsset, E., Ørka, H. O., Coops, N. C., ... &
Rombouts, C. (2012). Lidar sampling for large-area forest characterization: A review.
*Remote Sensing of Environment*, 121, 196-209.
https://doi.org/10.1016/j.rse.2012.01.020
Wulder, M. A., White, J. C., Varhola, A., Vastaranta, M., Coops, N. C., Cook, B. D., Pitt, D.,
& Woods, M. (2013). A best practices guide for generating forest inventory attributes
from airborne laser scanning data using an area-based approach. *The Forestry
Chronicle*, 89(6), 722-733. https://doi.org/10.5558/tfc2013-123
X AI. (n.d.). Grok [AI tool]. Retrieved from https://www.x.ai/grok
Xaxli’p Community Forest Corporation. (2009, July 6). Community forest agreement
application (Direct invitation to apply). Xaxli’p Community Forest Corporation.
Xaxli’p Community Forest Corporation. (2018). Forest stewardship plan 2018-2023:
Community forest agreement K3L. Xaxli’p Community Forest Corporation.
Xaxli’p Community Forest. (n.d.-a). Traditional use study. Retrieved July 28, 2024, from
https://www.xaxlipcommunityforest.ca/traditional-use-study
Xaxli’p Community Forest. (n.d.-b). Eco-cultural restoration. Retrieved July 28, 2024, from
https://www.xaxlipcommunityforest.ca/eco-cultural-restoration
Xaxli’p Government, & 3PIKAS. (2024). Xaxli’p land use plan (Draft). Xaxli’p
Government.
Xu, Y., Zhou, T., Zeng, J., Hui, L., Zhang, Y., Liu, X., … & Zhang, J. (2024). Spatial pattern
of forest age in China estimated by the fusion of multiscale information. Forests,
15(8), 1290. https://doi.org/10.3390/f15081290
Yao, Y., Zhao, X., Chang, L., Rong, J., Zhang, Y., Dong, Z., … & Su, Y. (2020). Vehicle fuel
consumption prediction method based on driving behavior data collected from
smartphones. Journal of Advanced Transportation, 2020, 1-11.
https://doi.org/10.1155/2020/9263605
Zhao, K., Popescu, S., & Nelson, R. (2011). ALS-based mapping of leaf area index and
structural parameters of pine stands. Remote Sensing of Environment, 115(10), 2741–
2750.
Zhao, K., Popescu, S., Meng, X., Pang, Y., & Agca, M. (2011). Characterizing forest canopy
structure with lidar composite metrics and machine learning. Remote Sensing of
Environment, 115(8), 1978-1996.
167

Zhu, W., Li, Y., Luan, K., Qiu, Z., He, N., Zhu, X., … & Zou, Z. (2024). Forest canopy
height retrieval and analysis using random forest model with multi-source remote
sensing integration. Sustainability, 16(5), 1735. https://doi.org/10.3390/su16051735

168

Appendix
Appendix 1
Calculation of Key Variables
In this study, we focused on several key variables to analyze forest fuel characteristics
and their implications for wildfire management. These variables were calculated using R
based on empirical data collected from large sample plots and allometric equations from
Ung, Bernier, and Guo (2008; Can J. For. Res. 38:1123-1132). Below is a summary of how
each key variable was calculated:
1. Stems per Hectare (stemsph): Stems per hectare were calculated by first summing the
tree count for each plot and then converting this value to a per-hectare basis using the
per hectare factor (phf). The tree count only included trees with a diameter at breast
height (DBH) greater than 10 cm.
o

Variable: stemsph

o

Formula: stemsph = tree.count * phf

o

Components:
▪

tree.count: The total number of trees in the plot.

▪

phf: Per hectare factor, calculated as phf = 1 / ((11.28^2 * pi) / 10000)
to convert plot area (400 m²) to hectares.

2. Canopy Fuel Load (CFL): The canopy fuel load represents the biomass in the tree
crowns. It was calculated by dividing the crown biomass per plot by the plot area
(400 m²). The biomass components were computed using species-specific allometric
equations derived from Ung, Bernier, and Guo (2008).
o

Variable: CFL

o

Formula: CFL = crown.bm.plot / (11.28^2 * pi)

o

Components:
▪

crown.bm.plot: Total crown biomass within the plot, calculated using
species-specific allometric equations for foliage and branches.

3. Canopy Bulk Density (CBD): The canopy bulk density was calculated by dividing the
crown biomass per plot by the plot area and then dividing by the crown depth. Crown
169

depth was determined as the difference between the 90th percentile tree height and
the median canopy base height.
o

Variable: CBD

o

Formula: CBD = (crown.bm.plot / (11.28^2 * pi)) / crown.depth

o

Components:
▪

crown.bm.plot: Total crown biomass within the plot.

▪

11.28^2 * pi: Area of the plot in square meters.

▪

crown.depth: Vertical distance between the top of the canopy and the
base of the live canopy, calculated as Q90.th - Med.CBH.

4. Understory, Ladder, and Canopy Fuel Loads: Fuel loads were divided into three
categories based on height: understory fuel (0-2 meters), ladder fuel (2 – median
canopy base height), and canopy fuel (median canopy base height and above). These
categories were determined using a custom function that summed the biomass in
specified height bins.
o

Variables: understory.fuel, ladder.fuel, crown.fuel

o

Calculation:

o

▪

Understory Fuel: understory.fuel = sum(canopy.bins[0:2])

▪

Ladder Fuel: ladder.fuel = sum(canopy.bins[2:canopy.base.bin])

▪

Canopy Fuel: crown.fuel =
sum(canopy.bins[canopy.base.bin:length(canopy.bins)])

Components:
▪

canopy.base.bin: Height at the base of the live canopy, rounded to the
nearest whole meter.

▪

canopy.bins: Binned biomass data calculated for each meter of tree
height.

5. Average and Median Heights: The average and median tree heights were calculated
by taking the mean and median of all tree heights (th) within each plot.
o

Variables: Avg_ht, max.th

o

Calculation:
▪

Average Height: Avg_ht = mean(th, na.rm = TRUE)

▪

Maximum Height: max.th = max(th, na.rm = TRUE)

170

6. Median Canopy Base Height (Med.CBH): The median canopy base height represents
the median height at which the live canopy begins.
o

Variable: Med.CBH

o

Calculation: Med.CBH = median(blc, na.rm = TRUE)

7. Total Biomass: Total biomass was calculated by summing the biomass of wood, bark,
foliage, and branches for each plot. The biomass components were computed using
species-specific allometric equations derived from Ung, Bernier, and Guo (2008).
o

Variable: total.biomass.plot

o

Formula: total.biomass.plot = sum(total, na.rm = TRUE)

o

Components:
▪

total: Total biomass of each tree, calculated using species-specific
allometric equations:
▪

Wood Biomass: ywood(D, H, bwood1, bwood2, bwood3)

▪

Bark Biomass: ybark(D, H, bbark1, bbark2, bbark3)

▪

Foliage Biomass: yfoliage(D, H, bfoliage1, bfoliage2,
bfoliage3)

▪

Branch Biomass: ybranches(D, H, bbranches1, bbranches2,
bbranches3)

▪

Crown Biomass: ycrown(yfoliage, ybranches)

▪

Total Biomass: ytotal(ywood, ybark, yfoliage, ybranches)

8. Basal Area: Basal area is the cross-sectional area of a tree trunk at breast height and is
expressed per unit area of land. It provides an estimate of tree density and stand
structure.
•

Variables: ba.per.ha, ba.plot

•

Calculation:
o

Basal Area per Plot: ba.plot = sum((((dbh)/2)^2*pi)/100, na.rm = TRUE)
▪

o

dbh: Diameter at breast height.

Basal Area per Hectare: ba.per.ha = ba.plot * phf
▪

phf: Per hectare factor, calculated as phf = 1 / ((11.28^2 * pi) /
10000).
171

9. Leaf Area Density (LAD): Leaf area density represents the amount of leaf area per
unit volume of canopy space. It is calculated in height bins, providing a detailed
vertical profile of the canopy's leaf area distribution.
1.

Variables: lad1, lad2, ..., lad40 (each representing leaf area density in 1-meter
height bins from 0 to 40 meters)

2.

Calculation: Leaf area density (LAD) was assessed by partitioning the point cloud
data into 1-meter horizontal slices using a standard extinction coefficient for
foliage (k = 0.5). The function leaf_area_density_bins calculated LAD for each
height bin by summing the leaf area density values for all horizontal slices within
the specified height range.
o

For each plot, LAD was computed as follows:
▪

The point cloud data (z) was partitioned into 1-meter slices.

▪

An extinction coefficient (k = 0.5) was used to account for foliage
distribution.

▪

The minimum height (z0) considered for leaf area density was set
to 0.01 meters.

▪

The leaf area density for each 1-meter height bin from 0 to 40
meters was calculated and recorded as lad1 to lad40.

Appendix 2
1. ALS Metric Variable Definitions
Standard Metrics (56):
area: The total area covered by the point cloud data.
ikurt: Kurtosis of the intensity values.
imax: Maximum intensity value.
imean: Mean intensity value.
ipcumzq10: 10th percentile of cumulative intensity.
ipcumzq30: 30th percentile of cumulative intensity.
ipcumzq50: 50th percentile of cumulative intensity.
172

ipcumzq70: 70th percentile of cumulative intensity.
ipcumzq90: 90th percentile of cumulative intensity.
ipground: Intensity of ground returns.
isd: Standard deviation of intensity values.
iskew: Skewness of intensity values.
itot: Total intensity.
n: Number of points.
p1th-p5th: Proportion of points in the 1st to 5th height bins.
pground: Proportion of ground points.
pzabove2: Proportion of points above 2 meters.
pzabovezmean: Proportion of points above the mean height.
zentropy: Entropy of height distribution.
zkurt: Kurtosis of height distribution.
zmax: Maximum height.
zmean: Mean height.
zpcum1-zpcum9: 1st to 9th percentile of cumulative height.
zq5-zq95: 5th to 95th percentile of height.
zsd: Standard deviation of height.
zskew: Skewness of height.
Custom Metrics (24):
Alscbh: Average lower stem count below a certain height.
Alscd: Average lower stem density.
Alsch: Average lower stem height.

173

Alscvlad: Average lower stem volume density.
isfd: Intensity standard deviation for the fuel distribution.
lad.canopy: Leaf area density in the canopy layer.
lad.ladder: Leaf area density in the ladder fuel layer.
lad.veg: Leaf area density in the vegetation layer.
lad 1-14: Leaf area density at height bin 1 - 14 meters.
VAI.AOP: Vertical area index for the area of interest.
VCI.AOP: Vertical canopy index for the area of interest.

2. Dependent Variables (Predicted Variables) at Landscape Scale for the Initial 2020
Sampling Campaign
Avg_blc: Average base of live canopy
Avg_cc: Average canopy circumference
Avg_dbh: Average diameter at breast height
Avg_ht: Average tree height
b1-m to b30-m: Biomass for metric bins 1 to 30 (0-30 meters) in kg/m²
b1-pha to b30-pha: Biomass per hectare for metric bins 1 to 30 (0-30 meters)
ba.per.ha: Basal area per hectare
ba.plot: Basal area per plot
bark.bm.pha: Bark biomass per hectare
bark.bm.plot: Bark biomass per plot
branch.bm.pha: Branch biomass per hectare
branch.bm.plot: Branch biomass per plot
CBD: Canopy bulk density
174

CFL: Canopy fuel load
CFL.pha: Canopy fuel load per hectare
crown.bm.pha: Crown biomass per hectare
crown.bm.plot: Crown biomass per plot
crown.depth: Crown depth
crown.fuel: Crown fuel load
cwd_sum: Coarse woody debris sum
cwd.volume_ha: Coarse woody debris volume per hectare
elevation_bins: Elevation in bins
elevation_m: Elevation in meters
foliage.bm.pha: Foliage biomass per hectare
foliage.bm.plot: Foliage biomass per plot
FuelCode: Fuel code classification
FuelDenMid: Fuel density in the mid-story based on visual estimate gathered at N,W, S, E
plot edges
FuelDenTop: Fuel density in the top-story based on visual estimate gathered at N,W, S, E
plot edges
FuelDenUnder: Fuel density in the understory based on visual estimate gathered at N,W, S, E
plot edges
InterSup_CBD: Intermediate support canopy bulk density
InterSup_stempsh: Intermediate support stems per hectare
ladder.fuel: Ladder fuel load
max.th: Maximum tree height

175

MDecay: Median decay class
Med.CBH: Median canopy base height
MidS: Mid-story biomass (Ladder fuel load, estimated for 1.37 to ~3 meters)
Q10.th: 10th percentile tree height
Q90.th: 90th percentile tree height
stem.bm.pha: Stem biomass per hectare
stem.bm.plot: Stem biomass per plot
stempsh: Stems per hectare
TopS: Top-story biomass (Canopy fuel load, estimated for above ~3 meters)
total.biomass.pha: Total biomass per hectare
total.biomass.plot: Total biomass per plot
tree.count: Tree count based on all trees measured in a large sample plot for the initial data
collection campaign (only trees larger than 10 cm DBH were measured)
UnderS: Understory biomass (Understory fuel load, estimated for 0-1.37 meters)
understory.fuel: Understory fuel load
wood.bm.pha: Wood biomass per hectare
wood.bm.plot: Wood biomass per plot

3. ALS Metrics Used for Model Training (Predictor Variables)
Standard Metrics (56): area, ikurt, imax, imean, ipcumzq10, ipcumzq30, ipcumzq50,
ipcumzq70, ipcumzq90, ipground, isd, iskew, itot, n, p1th, p2th, p3th, p4th, p5th, pground,
pzabove2, pzabovezmean, zentropy, zkurt, zmax, zmean, zpcum1, zpcum2, zpcum3, zpcum4,
zpcum5, zpcum6, zpcum7, zpcum8, zpcum9, zq5, zq10, zq15, zq20, zq25, zq30, zq35, zq40,
zq45, zq50, zq55, zq60, zq65, zq70, zq75, zq80, zq85, zq90, zq95, zsd, zskew
176

Custom Metrics (24): Alscbh, Alscd, Alsch, Alscvlad, isfd, lad.canopy, lad.ladder,
lad.veg, lad1, lad2, lad3, lad4, lad5, lad6, lad7, lad8, lad9, lad10, lad11, lad12, lad13, lad14,
VAI.AOP, VCI.AOP

4. Key Variables Calculated Dependent/Response Variables
CBD - Canopy Bulk Density
CFL - Canopy Fuel Load
stemsph - Stems per Hectare
crown.fuel - Canopy Fuel Load
ladder.fuel - Ladder Fuel Load
understory.fuel - Understory Fuel Load
TopS - Top-story biomass (Canopy fuel load, estimated for above ~3 meters)
MidS - Mid-story biomass (Ladder fuel load, estimated for 1.37 to ~3 meters)
UnderS - Understory biomass (Understory fuel load, estimated for 0-1.37 meters)
max.th - Maximum Tree Height
Avg_ht - Average Tree Height
Med.CBH - Median Canopy Base Height
total.biomass.plot - Total Biomass per Plot
ba.per.ha and ba.plot - Basal Area per hectare and per plot
lad1 to lad40 - Leaf Area Density in Height Bins

177

Appendix 3
Variable Importance Plots
Understory Fuel:

The most important predictors for understory fuel include total intensity of LiDAR
returns (itot), cumulative percentage of returns above the 90th height quantile (ipcumzq90),
leaf area density at the 5th height quantile (lad5), and skewness of the height distribution
(zskew). These metrics, while influential in the model, may seem counterintuitive. For
instance, upper canopy metrics like ipcumzq90 significantly affect understory predictions,
likely due to the indirect influence of canopy structure on the understory environment (e.g.,
light penetration and microclimatic conditions).

178

Ladder Fuel:

Key predictors for ladder fuel include the 35th height quantile (zq35), leaf area
density in the canopy layer (lad.canopy), skewness of the height distribution (zskew), and
leaf area density at the 14th height quantile (lad14). These predictors emphasize the vertical
continuity of vegetation, suggesting that managing these layers can reduce the risk of fire
spreading from the ground to the canopy.

179

Crown Fuel:

For crown fuel, the most critical predictors are the 35th height quantile (zq35), 30th
height quantile (zq30), standard deviation of height (zsd), and skewness of the height
distribution (zskew). These metrics indicate that accurate mapping of crown fuel benefits
from detailed LiDAR data capturing the canopy structure. Management strategies should
consider the spatial distribution and density of mid-canopy vegetation.

180

Canopy Bulk Density (CBD):

The top predictors for CBD include the 35th height quantile (zq35), proportion of
returns above the mean height (pzabovezmean), skewness of the height distribution (zskew),
and cumulative percentage of returns above the 50th height quantile (ipcumzq50). These
predictors highlight the importance of canopy density and vertical structure, suggesting that
effective management should aim to modify these attributes to reduce fire intensity and
improve fire resistance in forested areas.
Overall, these findings provide valuable insights into the factors influencing forest
fuel characteristics and can guide future research and operational efforts in wildfire risk
mitigation and planning. For those interested in the technical details, the variable importance
plots and their interpretations are included in the appendix.

181

Predicted vs. Observed Plots:
Understory Fuel:

182

Ladder Fuel:

Crown Fuel:

183

Canopy Bulk Density (CBD):

Median Canopy Base Height (CBH):

184