Comparison of methods to estimate fuel moisture content in different forest stand
types in central British Columbia
By
Rulan Xiao
B.Sc., Simon Fraser University, 2020
A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR
THE DEGREE OF MASTER OF SCIENCE IN
The Natural Resources and Environmental Studies Graduate Program
(Geography)

Supervisor: Dr. Joseph Michael Shea
Committee members: Drs. Phil Burton and Ché Elkin

UNIVERSITY OF NORTHERN BRITISH COLUMBIA
May 2023

© Rulan Xiao, 2023

2

Abstract
Wildfires are a growing threat due to climate change, and they often leave
unburned forest patches called fire refugia. While young forests in some regions burn
more severely, preliminary observations in central British Columbia suggest that
managed juvenile forests exhibit lower fire severity, potentially influenced by fuel
moisture conditions, and stand characteristics. To identify the role of fuel moisture in the
formation of juvenile stand fire refugia, this research collects and examines groundbased, empirically modelled, and remote sensing indices of greenness, moisture, and
fire severity.
This thesis investigates the fuel moisture contents (FMC) of duff, fine woody
debris, and foliage at six locations near Prince George and Smithers, British Columbia
(BC), over two summers (2021 and 2022). A total of 6116 individual samples of foliage,
fine woody debris, and duff were collected from open, juvenile, and mature conifer
forest stands and analysed for moisture content (MC). On average, the MC of duff and
fine woody debris samples was higher in juvenile and mature forests than open sites. In
contrast, open forests had higher foliage MC than the other forests. Observations of
FMC were used to evaluate the accuracy of FMC estimates extracted from the
Canadian Forest Fire Weather Index (FWI) system. Observations of FMC were also
compared with remote sensing indices to assess the utility of using spaceborne
(Landsat 8&9, Sentinel 2) remote sensing to predict local FMC.
Three versions of the FWI model were used to estimate FMC: the original FWI
model which uses the closest fire weather station, and versions that used updated
parameters based on local fuel conditions and in-stand weather data. When estimating
fine woody debris MC, the best statistical results are obtained with locally calibrated
models at open stands. However, the original FWI model provides better estimates of
duff MC in juvenile stands.
For remote sensing of foliar MC in juvenile stands, the Normalized Difference
Moisture Index (NDMI) had a higher R2 value (0.334) and a lower RMSE than other
indices, while the NDMI gave the best result for foliar MC in mature forests (R2 = 0.160).
For fine woody debris and duff MC in open stands, none of the remote sensing indices

3
tested have R2 > 0.1 when estimating duff and fine woody debris MC. However, the
RMSE of using empirical models from FWI to estimate duff (lowest at 59.95% RMSE)
and fine woody debris (21.13%) MC was higher than remote sensing (41.64% for duff,
17.50% for fine woody debris).
Remote sensing indices such as NDMI and GNDVI (Green Normalized
Difference Vegetation Index) were used to estimate pre-burn FMC, and the estimated
FMC results were found to be generally higher at juvenile stands than mature forest
from a case study area from Plateau Complex Wildfire of 2017. Lower remote sensing
estimates of FMC in mature stands corresponded to higher burn severities.

4

Acknowledgments
I like to take this opportunity to express my gratitude to Dr. Joseph Shea, my
supervisor, for providing me with the chance to work on such an important project. This
study is a side extension of Dr. Phil Burton's project, which already had a good quantity
of material for me to work with. Dr. Phil Burton and Dr. Ché Elkin, members of my
committee, have provided numerous helpful comments regarding my thesis work
process. Vanessa Foord provided and installed weather sensors for this research,
which was extremely important for data collection.
Mackenzie McLean, Morgan Endacott, Dongxu Piao, Shiyuan Jing, Allie Golt at
Prince George, and Clayton Rose and Trevor Schibli at Smithers have been invaluable
in gathering fuel moisture data for the 2021 and 2022 summers.
My funding comes from the Natural Sciences and Engineering Research Council
and Natural Resource Canada. Also, UNBC provided me with teaching assistant
opportunities as well, which also helped this research proceed. My parents Shu Cong
and Ruiqi Xiao have been understanding and supportive of my ongoing research
journey.
I acknowledge my work is located on unceded Lheidli T'enneh and Wet’suwet’en
territory.

5

Table of contents
Abstract

2

Acknowledgments

4

1. Introduction

12

1.2 Thesis statement
2. Literature review

14
14

2.1 Field Observations of Fuel Moisture Content

16

2.2 Fuel moisture inferred from weather station data

17

2.2.1 Fine Fuel Moisture Code (FFMC)

18

2.2.2 Duff Moisture Code (DMC)

19

2.3. FMC from remote sensing
3. Study Areas

21
24

3.1 Prince George and Smithers

24

3.2. Plateau Complex Wildfire, Williams Lake

28

4. Methods
4.1 Fuel Moisture Observations

32
32

4.1.1 Sample collection

32

4.1.2 Moisture content measurement

33

4.1.3 Moisture content analysis

35

4.2 Regional and on-site meteorological data

35

4.2.1 Fire weather stations

35

4.2.2 In-stand meteorological observations

36

4.2.3 Meteorological data analyses

36

4.3 FWI empirical model
4.3.1 Model 1: Fire weather stations and empirical coefficients

37
37

6
4.3.2 Model 2: FFMC and DMC corrected for local moisture contents.

38

4.3.3 Model 3: FWI with local temperature and relative humidity

41

4.3.4 Model 4: FWI with local weather conditions

42

4.4 Remote sensing and GIS

42

4.4.1 Remote Sensing Indices For FMC

43

4.3.2 Mapping Forest types with GIS

47

4.3.3 Relations between dNBR/RBR and remote sensing indices

48

5. Results

50

5.1 Fuel Moisture Contents

50

5.2 Meteorological Data

58

5.3 Estimates of FMC from FWI

62

5.3.1 Model 1: Original model with fire weather stations and empirical
coefficients
5.3.2 Model 2: FFMC and DMC correction for local MC

62
64

5.3.3 Model 3: FFMC and DMC with local temperature and relative humidity
67
5.3.4: Model 4: FFMC and DMC with local weather conditions
5.4 Estimates of FMC from Remote Sensing

70
72

5.4.1 Estimating FMC with Sentinel 2 data in 2021

72

5.4.2 Estimating FMC with Landsat 8 & 9 data in 2022.

78

5.4.3 Estimating FMC with Sentinel 2 data in 2022.

83

5.5 Burn severity, moisture, indices, and FMC
6. Discussion

88
102

6.1 Field observations

102

6.2 Empirical Models to Estimate Fuel Moisture Content

103

6.3 Remote Sensing of FMC

105

7
6.4 Burn Severity Case Study

107

7. Conclusions

108

8. References

110

Appendix

116

Figures:
Figure 1:Schematic of FWI, modified from (Natural Resources Canada,
accessed May 2023) ..................................................................................................... 17
Figure 2: Prince George field sites (imagery from Google EarthTM) .................... 26
Figure 3: Smithers field sites (Image clipped: Google Earth TM) .......................... 27
Figure 4: Tamarac sampling locations near the BCWS Bednesti fire weather
station, showing three stand types (Image clipped: Google EarthTM, RPAS image:
Dr.Joseph Shea) ........................................................................................................... 28
Figure 5: Plateau Complex Wildfire map and study area (image captured in
Google EarthTM). ........................................................................................................... 30
Figure 6: False colour composite pre-fire map at Plateau Complex, image
captured on June 11th, 2017, centred on 52.94105 N, -124.01571 W. ......................... 31
Figure 7: False colour composite postfire map at Plateau Complex, image
captured on September 4th, 2017, centred on 52.94105 N, -124.01571 W. ................. 32
Figure 8: Drying process at UNBC EFL lab ........................................................ 34
Figure 9: FFMC and DMC application flow chart ............................................... 38
Figure 10: Tamarac NDVI in different stands in 30 metres pixels in three stand
types, NDVI captured on July 20th, 2022. ...................................................................... 45
Figure 11: Example of lowess smoothing of NDMI, at Tamarac juvenile site in
2022 summer ................................................................................................................ 46
Figure 12: Stand ages with true colour image on 6th July 2017, centred on
52.94105 N, -124.01571 W. .......................................................................................... 48
Figure 13: Daily average FMC in 2022 for different stands and fuel types at the
Prince George and Smithers. ........................................................................................ 53

8
Figure 14: Boxplots of daily averaged duff MC at different stands and locations in
2021 and 2022 around Prince George and Smithers. ................................................... 55
Figure 15: Boxplots of woody debris MC at different stands and sample collection
sites in 2021 and 2022 around Prince George and Smithers. ....................................... 56
Figure 16: Boxplots of foliage MC at different stands and sample collection sites
in 2021 and 2022 around Prince George and Smithers. ............................................... 57
Figure 17: Weather condition comparison between the BC fire weather station
(FWS, Houston) and onsite weather station in 2022 at Barren, BC. ............................. 59
Figure 18: Weather condition comparison between on-site weather stations at
different sites in 2022 at Barren, BC.............................................................................. 61
Figure 19: Estimated vs observed FMC from as-is weather empirical model at
different stands .............................................................................................................. 63
Figure 20: Estimated versus observed FMC from FWI (original model corrected
by local MC). ................................................................................................................. 65
Figure 21: Estimated versus observed FMC with local in-stand temperature and
relative humidity ............................................................................................................ 68
Figure 22: Estimate vs observation FMC with local weather conditions. ............ 71
Figure 23: Example of remote sensing indices. A) NMDI, B) VARI, C) GNDVI, D)
NDMI, E) NDVI at North Fraser sites on June 2nd , 2022 .............................................. 74
Figure 24: Foliage MC compared with five remote sensing indices at juvenile
stands with Sentinel 2 data in 2021. .............................................................................. 75
Figure 25: Foliage MC compared with five remote sensing indices at mature
stands with Sentinel 2 data in 2021 ............................................................................... 75
Figure 26: Duff MC compared with five remote sensing indices at open stands
with Sentinel 2 data in 2021. ......................................................................................... 76
Figure 27: Fine woody debris MC compared with five remote sensing indices at
open stands with Sentinel 2 data in 2021. ..................................................................... 76
Figure 28: Foliage MC compared with five remote sensing indices at juvenile
stands with Landsat 8 & 9 data in 2022 ......................................................................... 79
Figure 29: Foliage MC compared with five remote sensing indices at mature
stands with Landsat 8 & 9 data in 2022......................................................................... 79

9
Figure 30: Duff MC compared with five remote sensing indices at open stands
with Landsat 8 & 9 data in 2022. ................................................................................... 81
Figure 31: Fine woody debris MC compared with five remote sensing indices at
open stands with Landsat 8 & 9 data in 2022 ................................................................ 81
Figure 32: Remote sensing indices compared with foliage MC for juvenile stands
with Sentinel 2 data in 2022 .......................................................................................... 83
Figure 33: Remote sensing indices compared with foliage MC for mature stands
with Sentinel 2 data in 2022. ......................................................................................... 84
Figure 34: Remote sensing indices compared with duff MC at open stands with
Sentinel 2 data in 2022.................................................................................................. 86
Figure 35: Remote sensing indices compared with woody debris MC at open
stands with Sentinel 2 data in 2022. .............................................................................. 86
Figure 36: dNBR for different stands on 6th July 2017 at Plateau Complex,
centred on 52.94105 N, -124.01571 W. ........................................................................ 89
Figure 37: RBR value for different stands on 6th July 2017 at Plateau Complex,
centred on 52.94105 N, -124.01571 W. ........................................................................ 90
Figure 38: Histogram of dNBR for different stand types, where dashed lines
represent median dNBR. ............................................................................................... 91
Figure 39: RBR histogram in different stands, where dashed lines represent
median RBR. ................................................................................................................. 91
Figure 40: A) NDVI, B) NDMI, C) NMDI, D) VARI and E) GNDVI vs dNBR for
mature forests, with their frequency distributions .......................................................... 92
Figure 41: A) NDVI, B) NDMI, C) NMDI, D) VARI and E) GNDVI vs dNBR for
open forests, with their frequency distributions ............................................................. 93
Figure 42: A) NDVI, B) NDMI, C) NMDI, D) VARI and E) GNDVI vs dNBR for
juvenile forests, with their frequency distributions. ........................................................ 94
Figure 43: A) NDVI, B) NDMI, C) NMDI, D) VARI and E) GNDVI vs RBR for
mature forests, with their frequency distributions. ......................................................... 95
Figure 44: A) NDVI, B) NDMI, C) NMDI, D) VARI and E) GNDVI vs RBR for open
forests, with their frequency distributions. ..................................................................... 96

10
Figure 45: A) NDVI, B) NDMI, C) NMDI, D) VARI and E) GNDVI vs RBR for
juvenile forests, with their frequency distributions. ........................................................ 97
Figure 46: FMC calculated by NDMI for juvenile and mature stand on 6th July
2017 at Plateau Complex, centred on 52.94105 N, -124.01571 W. .............................. 98
Figure 47: FMC calculated by GNDVI at juvenile and mature stand on 6th July
2017 at Plateau Complex, centred on 52.94105 N, -124.01571 W. .............................. 99
Figure 48: A). FMC calculated by NDMI at mature stands, B). FMC calculated by
GNDVI at mature stands, C). FMC calculated by NDMI at juvenile stands, D). FMC
calculated by GNDVI at juvenile stands compared with dNBR. ................................... 100
Figure 49: A). FMC calculated by NDMI at mature stands, B). FMC calculated by
GNDVI at mature stands, C). FMC calculated by NDMI at juvenile stands, D). FMC
calculated by GNDVI at juvenile stands compared with RBR. ..................................... 101
Figure 50: Usable NDMI indices in 6 sample collection sites in 2021 summer (late
May to September) ...................................................................................................... 106
Tables:
Table 1:Satellite sensors used in this study ........................................................ 22
Table 2: Band indices used in this study ............................................................ 23
Table 3: Initial FMC at different sites and stand around Prince George and
Smithers in 2021 and 2022. .......................................................................................... 39
Table 4: Burn severity levels obtained calculating dNBR ................................... 49
Table 5: Burn severity levels obtained calculating RBR. .................................... 49
Table 6: The maximum, minimum of daily averaged MC, and averaged of daily
averaged MC for duff and woody debris from different stands in 2021 and 2022. ........ 51
Table 7: The maximum, minimum of daily averaged MC, and averaged of daily
averaged MC for foliage from different stands around Smithers and Prince George in
2021 and 2022 .............................................................................................................. 51
Table 8: The maximum, minimum of daily averaged MC, and averaged of daily
averaged MC for foliage from different stands in 2021 and 2022. ................................. 52
Table 9: Errors in estimated MC for different stand and fuel types for the ‘as-is’
empirical model (FFMC/DMC) in 2021 and 2022. ......................................................... 64

11
Table 10: Errors in estimated MC for different stand and fuel types for weather
empirical model (FFMC/DMC) at different stands in 2021 and 2022, corrected by local
MC................................................................................................................................. 66
Table 11: Errors in estimated MC for different stand and fuel types for the
weather empirical model (FFMC/DMC) calibrated with local temperature and relative
humidity, as well as local fuel correction. ...................................................................... 69
Table 12: Errors in estimated MC for different stand and fuel types for weather
empirical model (FFMC/DMC) with local temperature, relative humidity, wind speed, and
accumulated precipitation, as well as local fuel correction. ........................................... 71
Table 13: Study area’s cloud/shadow free images in the summer of 2021 and
2022 from Smithers and Prince George ........................................................................ 72
Table 14: Statistical results of remote sensing indices with second year foliage
MC at juvenile and mature stands (2021 Sentinel 2) ..................................................... 77
Table 15: Statistical results of remote sensing indices with fine woody debris/ duff
MC at open stands (2021 Sentinel 2). ........................................................................... 78
Table 16: Statistical results of remote sensing indices with foliage MC at juvenile
and mature stands (2022 Landsat 8 & 9). ..................................................................... 80
Table 17: Statistical results of remote sensing indices with duff and fine woody
debris MC at open stands (2022 Landsat 8 & 9) ........................................................... 82
Table 18: Statistical results of remote sensing indices relationships with foliage
MC at juvenile and mature stands (Sentinel 2, 2022). ................................................... 84
Table 19: Statistical results of remote sensing index relationships with fine woody
debris and duff MC at open stands (2022 Sentinel 2) ................................................... 87
Appendix:
Appendix 1: Sample sites around Prince George and Smithers ....................... 116
Appendix 2: Sample sites around Prince George and Smithers ....................... 117
Appendix 3: List of weather stations installed and used in this research in 20212022. ........................................................................................................................... 118
Appendix 4: List of weather stations installed and used in this research in 20212022 ............................................................................................................................ 119

12
Appendix 5: Average and variance of temperature (T, °C), relative humidity
(RH, %), wind speed (ws, kph) as well as Precipitation (P, mm) in different stands from
data collected in summer of 2022. .............................................................................. 120
Appendix 6: A) Cloud and shadow-free true colour image, and B) NDVI for the
North Fraser site, image captured on June 2nd, 2022, centred at 54.252924 N,
122.499171 W. ............................................................................................................ 121
Appendix 7: -A Cloudy and cloud-shadowed true colour image, and B) NDVI for
the North Fraser site, image captured on June 4th, 2022, centred at 54.252924 N,
122.499171 W. ............................................................................................................ 122
Appendix 8: A) Cloud true colour image, and B) NDVI for the North Fraser site,
image captured on June 9th, 2022, centred at 54.252924 N, 122.499171 W. ............ 123
Appendix 9: Daily averaged FMC in 2021 for different stands and fuel types at
the Prince George and Smithers study locations. ....................................................... 124

1. Introduction
Wildfires are a serious threat to humans, infrastructure, and industry (McGee et
al., 2015). They are expected to increase in frequency and severity as a result of
climate change (Flannigan et al., 2000). In the western Cordillera of North America,
wildfires frequently leave unburned blocks of forest known as fire refugia or fire islands
(Krawchuk et al., 2016). According to a prior study in southwestern Oregon, USA, young
forests burn more severely, and many old forests were left largely unburned as wildfire
refugia (Zald & Dunn, 2018). However, preliminary observations from central British
Columbia (BC) showed that some juvenile forests burn less severely than mature
forests (Burton et al., 2019). Typically, distinct from naturally regenerating forests,
young forests planted after logging in central BC are characterised by a dense uniform
structure of a single tree species (Wang et al., 2019). Managed juvenile forest stands in
central BC have several plantation features: uniform tree species (often lodgepole pine,
Pinus contorta var. latifolia trees) and tight spacing between them (Burton, 2022). The
difference between fire refugia in central BC and other regions could be caused by a

13
disparity in fuel moisture conditions, the effects of the stand characteristics on the
factors that influence fire spread, or both. Previous research has shown that fuel
components in different stand types can influence fire behaviour (Kane et al., 2015;
Dunn & Bailey, 2016).
Specific approaches have used remote sensing and weather data to estimate the
fuel moisture content (FMC) of flammable materials found in nature. Fuel moisture
content is a strong determinant of wildfire susceptibility and behaviour (Scott & Burgan,
2005, Krawchuk et al. 2009). The Canadian Forest Fire Weather Index (FWI) system is
widely used to measure and communicate forest fire danger in Canada (Van Wagner,
1987). Parts of FWI can be used to estimate the moisture content (MC) of fine woody
debris and duff around fire weather stations, which were located in forest openings (Van
Wagner, 1987; De & William, 1998). Previous research has used remote sensing to
estimate MC in broadleaf (Danson & Bowyer, 2004) and coniferous foliage in western
Montana, USA (Qi et al., 2014). Previous studies have also used remotely sensed burn
severity indices to help understand fire behaviour from historical fires (Parks et al.,
2014). Due to the challenge of diverse, complex and interacting fire effects, there is a
limitation in using remote sensing indices to assess burn severity (Penelope et al.,
2014). However, there needs to be more research to accurately estimate FMC in
different forest stands. It is necessary to compare methods to estimate selected FMC
attributes in central BC and determine whether differences in FMC might explain why
juvenile forests sometimes remain as wildfire refugia in central BC, in contrast to
patterns observed in southwestern Oregon (Zald & Dunn, 2018).
This study includes ground-based observations of fuel moisture levels against
directly

measured

weather

conditions,

standard

empirically

based

modelling

approaches, and remote sensing data. Weather conditions, empirical models and
remote sensing methods are tested to estimate FMC in different forest stands.
Estimated FMC from these alternative data sources were also used to analyse
observed burn severity differences in the Plateau Complex Wildfire in central BC in
2017. Pre-fire evaluation and prevention could be accomplished by using the methods
determined by this research, with monitoring and assessing FMC on a larger scale for
different stand types (Chuvieco et al., 2003).

14

1.2 Thesis statement
There are two statements I aimed to prove in this research: 1. Empirical models
and remote sensing can be used to estimate fuel moisture contents on a larger scale
(Central BC) 2. High fuel moisture contents played a role in juvenile stand fire skips in a
recent wildfire. To accomplish the research stated above, this study has:
1) Developed and tested empirical models of the FWI system to estimate duff
and fine woody debris MC from regional weather conditions and on-site
fire weather station data with field observation data.
2) Collected and analysed remote sensing imagery, calculated greenness
and MC indices, and compared them with field observations and
empirically modelled FMC values.
3) Compared burn severities calculated with remote sensing imagery for a
historical wildfire with remote sensing indices related to FMC and
evaluated the role of FMC in fire severity and fire refugia.

2. Literature review
In Canada, wildfires consume more than two million hectares of forest yearly
(McGee et al., 2015). Fires have an impact on forested areas around the world and
contribute significantly to greenhouse gas emissions (Chuvieco, 2008). Many
ecosystems are well adapted to fire cycles, and fire has historically been used as a tool
for land use management (Chuvieco et al., 2010). However, recent changes in the
climate and societal factors can alter traditional fire regimes, potentially magnifying the
adverse effects of fire on vegetation, soils, and human values (Chuvieco et al.,
2010).Based on inflation-adjusted insurance claims, the most expensive wildfire
disasters in Canada include the Kelowna wildfires in British Columbia in 2003 (costing
$252 million), the Slave Lake wildfires in Alberta in 2011 ($864.67 million), the Horse
River Wildfire in northeastern Alberta in 2016 (the costliest natural disaster in Canadian
history, at $3.84 billion), and the British Columbia wildfires in 2017 ($137.3 million)
(Tymstra et al., 2020). When the costs of suppression, recovery, lost revenue, and other

15
effects (such as on the air and water quality) are also considered, the overall cost of
wildfire disasters listed above is significantly higher (Tymstra et al., 2020).
Wildfires are a common disaster with substantial consequences, necessitating
research on wildfire prevention and mitigation (Tymstra et al., 2020). The most
comprehensive indicator of the potential for fire ignition and spread is fuel moisture
content (FMC), which is expressed as the ratio of water to dry mass (Blackmore &
Flanner, 1968; Fosberg & Schroeder, 1971; Paltridge & Barber, 1988; Pompe & Vines,
1966; Trowbridge & Feller, 1988; Viegas et al., 1992). The FMC strongly impacts the
time to ignition and the manner of fire development (Nelson, 2001).
It has been historically observed that old forests often functioned as fire refugia
that experienced lower severity burns in stand-replacing fires (Meigs et al., 2020).
However, according to observations from recent fires in central BC, some juvenile forest
stands escaped wildfire while mature forest burned around them, a phenomenon that
occurred more frequently than could be expected by chance (Burton et al., 2019).
Studies in California and Oregon have demonstrated that the fuel components in
different stand types of influence fire intensity and behaviour (Kane et al., 2015; Dunn &
Bailey, 2016). While there are many factors that can influence fire behaviour, fuel
moisture content (FMC) is an important measure of moisture in forest fuels that can
have a significant impact (Anderson 1982, Jolly et al. 2015). Other factors that can
affect fire behaviour include differences in tree species, site preparation (e.g., soil types),
and the specific area, level, and stage of fire initiation and spread (Byram 1959, White
et al. 1996). Nonetheless, FMC remains a crucial variable in assessing fire risk and
developing effective fire management strategies (Anderson 1982, Jolly et al. 2015).
Field observations of FMC provide accurate observations on one fire risk factor
but are time-consuming and only represent conditions in the sample collection area at
specific points in time (Caccamo et al., 2011). The FWI system has been used to
estimate weather-driven fire risk and local fuel moisture contents such as fine woody
debris and the forest floor or duff (Van Wagner, 1987). The associated Canadian Forest
Fire Danger Rating System (CFFDRS) is widely used to measure and report forest fire
danger across Canada today (Wang et al., 2017). Unfortunately, FWI needs accurate
weather conditions from weather stations, which can limit its application in sparsely

16
instrumented and topographically complex landscapes (Van Wagner, 1987). Remote
sensing has been suggested as an alternative approach to estimating FMC on a larger
scale. However, satellite image resolution (20-200 m), quantity (image collection
frequency), and quality (cloud cover) are challenges in operationalizing this research
(Yang et al., 2018). The following literature review examines the principles and
background of the three main components of this study: ground-based FMC
observations, empirical modelling of FMC, and remote sensing for FMC and wildfire
severity.

2.1 Field Observations of Fuel Moisture Content
Duff (forest floor organic matter consisting of fermentation and humus layers,
typically at 5-10 cm below the surface) and fine woody debris (small dead fuel such as
twigs and branches with diameter of 2-7mm), play a role in wildfire intensity and spread
(Ryan, 2002, Norum et al., 1984). In regions where the ground cover has high duff and
woody debris composition, fires burn several hours longer than in regions with
widespread mineral soil exposure, and this increases the chances of fire spreading to
foliage (Ryan, 2002; Hartford & Frandsen, 1992).
Fuel moisture content can substantially affect ignition probability and the severity
of forest fires, as low fuel moisture levels can be necessary for wildfire initiation and
spread (Dennison et al., 2008; Chuvieco et al., 2009; Nolan et al., 2016). Fuel moisture
estimation is thus necessary for fire risk assessment and prevention (Dennison et al.,
2008; Yebera et al., 2013).
Fuel moisture content is calculated as the percent difference between a sample's
fresh weight and dry weight (Equation 1, Chuvieco et al. 2003):
=

∗ 100%

[Eq. 1]

where FW is the fresh weight, and DW is the dry weight after oven-drying at 100 ℃ for
at least 24 hours. The typical ranges of FMC vary with fuel type, forest age, and recent
weather conditions and events. In FWI models, the MC of fine woody debris can range
between 0 to 250 %, and the MC of duff is between 20 to 300 % (Van Wagner, 1987).
The FWI model is based on observations from boreal and eastern Canada dominated

17
by pine stands of various ages and structures, so fuel conditions may respond
differently to weather in central BC (Wagner, 1987).

2.2 Fuel moisture inferred from weather station data
The FWI system uses weather station data to estimate fire risk and FMC of fine
fuels and duff for the area adjacent to the index weather station (Van Wagner, 1987; De
& William, 1998). The system includes a Fine Fuel Moisture Code (FFMC) and a Duff
Moisture Code (DMC), which are empirically calibrated estimates of the cumulative fuel
drying/wetting effects of recent weather behaviour (Viegas et al., 2001). The FFMC and
DMC indices (Figure 1), which are unitless, can be used to convert back to the MC of
fine woody debris and duff, respectively, by using formulas based on an empirical model
or by creating local formulas (Wotton, 2009).

Figure 1:Schematic of FWI, modified from (Natural Resources Canada, accessed May
2023)
The calculation of FFMC requires inputs of accumulated daily precipitation (mm),
relative humidity (%), wind speed (km hr-1), and temperature (oC) at noon, and the
previous day’s value of FFMC (Van Wagner, 1987; Wotton, 2009). As the duff layer is
found beneath the surface of the forest floor, it is unaffected by wind. The calculation of

18
DMC requires daily precipitation (mm), relative humidity (%), temperature ( oC), and day
length (h) and the previous day’s DMC (Van Wagner, 1987; Wotton, 2009).

2.2.1 Fine Fuel Moisture Code (FFMC)
The Fine Fuel Moisture Code (FFMC) assesses the flammability and ease of
ignition of fine fuels, such as fine woody debris, which typically have a dry weight of
around 0.25 kg m-2 (Van Wagner, 1987). It estimates the MC of litter and other cured
fine fuels within a forest stand during mid - afternoon (Van Wagner, 1987). FFMC
calculations require input data of temperature, relative air humidity, wind speed at noon,
and precipitation from 24 hours measured at a nearby fire weather station. Wetting by
rain and drying is also considered in FFMC calculations. The FFMC index is calculated
from equations 2 to 11 as follows (Van Wagner & Pickett, 1985):
Calculation of the current day’s FFMC (FFMCt) requires the previous day’s (t-1)
moisture content (MC, %), which is a function of the previous days FFMCt-1:
=

. ∗

[Eq. 2]

.

Precipitation (P, mm) is measured from noon of the previous day to noon of the
current day. If the precipitation amount is greater than 0.5 mm, the wetting effect from
rain is added to the previous day’s fine fuel moisture content (MCt-1) to get today’s fine
fuel moisture content (MCt). The wetting effect is estimated empirically as follows:
=

+

. ∗

=

. `+ .

∗

.

−

(

−

.

)

, for

, for

<150

[Eq. 3]
[Eq. 4]

≥150

In the FFMC model, the maximum fine fuel MC is 250%. If a moisture content calculated
from Eq.3 or 4 is higher than 250%, then it is converted back to 250. Then a drying
effect (Ed) considers air temperature (T, °C) and relative humidity (H, %) measured at
noon on the current day, and is calculated as follows:
.

= .

+

∗

+ .

(

. − )∗( −

.

)

[Eq.5]

If Ed is less than MCt-1, the fine fuel moisture content MCt is calculated by including the
wind speed (W):
= .

−

.

+ .

.

−

.

.

[Eq.6]

19
+(

=

[Eq.7]

) ∗ 10

−

If Ed is greater than MCt-1 then Eq. 7 would lead to a negative moisture content. So
instead, a wetting effect Ew that considers temperature and humidity is used to calculate
MCt:
.

= .

+

+ .

(

. − )

−

.

∗

[Eq.8]

If Ew >MCt-1 , then MCt can be calculated as following equations:
=

−(

)

−

−

[Eq.9]

If Ew≤ MCt-1≤ Ed , then there is no drying effect on the previous day’s moisture content
and Mt-1 is equal to MCt . Finally, the current day’s moisture content MCt is converted
back to FFMC following:
=

.

[Eq. 10]

.

In areas where snow is typically present in the winter, the calculation of FFMC
begins on the third day after the disappearance of snow. When snow cover is not a
prominent characteristic in a region, the calculation begins on the third day in a row that
the noon temperature is over 12°C (Lawson and Armitage 2008). The FFMC index is
set to 85 as its initial value.

2.2.2 Duff Moisture Code (DMC)
Another fuel moisture code from the Canadian forest fire weather index (FWI) system is
the Duff Moisture Code (DMC). Duff consists of loosely packed, decomposing organic
matter in the fermentation and humus layers (Norum et al., 1984) of forest soils, and it
has a dry weight of around 5 kg m-2 (Van Wagner, 1987). Similar to FFMC, the
calculation of DMC considers the wetting and drying phases. Required inputs to the
DMC calculation are temperature, relative air humidity, daily precipitation (at noon), the
previous day’s DMC, and day length (Van Wagner, 1987). DMC is calculated from
equations 12 to 19 as follows (Van Wagner & Pickett, 1985):
Duff requires daily precipitation (P) greater than 1.5 mm to achieve a wetting effect;
rainfall effect (Pe) is calculated by:
= .

− .

, for P >1.5 mm

[Eq.11]

20
The wetting variable (b) for previous day’s DMC (DMCt-1) is calculated for three different
DMCt-1 ranges (less than 33, between 33 to 65, higher than 65):
=

for DMCt-1 ≤33

.

=

− .

= .

(

(

[Eq.12]

), for 33< (DMCt-1) ≤65

[Eq.13]

, for DMCt-1 >65

[Eq.14]

)−

.

Once wetting effect b is calculated, it can use the previous day’s duff moisture content
(MCduff t-1) to get the current day’s duff MC (

), first estimating MC

duff t-1 from DMCt-1 :

=

+

=

.

[Eq.15]

.

+

[Eq.16]

.

MCduff is then converted back to the current day’s DMC for the next day continuous
calculation:
=

.

−

.

(

- 20)

[Eq.17]

As DMC can not go below 0, if DMC from Eq.18 is lower than 0, it will be set to 0 for the
continuous calculation. If P is less than 1.5 mm, the above calculation is not used and
DMC = DMCt-1. After calculating the DMC with the rainfall effect, a drying effect will also
be added to get the final DMC (DMCf ) for the current day as follows:
=

+

∗( .

∗( + . )∗(

−

)∗

∗

)

[Eq.18]

where T is temperature, H is relative humidity measured at noon, and DL is the day
length; notice the minimum temperature is set as -1.1 ℃, and any inputs lower than -1.1℃
will be set as -1.1℃. The DMC computation begins on the third snow-free day in areas
typically covered in snow throughout the winter. When snow cover is not prominent in a
region, the calculation begins on the third day in a row that the noon temperature is over
12 °C (Lawson & Armitage, 2008). The index's initial value must be set to 6.
Though the fine fuels and duff moisture codes can be used to estimate the MC of
woody debris and duff layers, the accuracy will vary with the distance between the site
and the fire weather stations, as temperature, relative humidity, wind speed, and
precipitation will vary. The drying and wetting corrections for FFMC and DMC are
empirical corrections based on data collected from major eastern pine forests around
Ontario (Van Wagner, 1987). In contrast, fuel conditions might differ in other regions

21
and forest types. Fire weather stations are usually located in open spaces close to a
highway. Previous research has used FFMC and DMC in old-growth forests in Australia,
with in-stand weather stations, where it was shown that the microclimate is moister in
the mature forest than in open and younger forests (Furlaud et al., 2021).

2.3. Fuel moisture from remote sensing
Remote sensing and spectral reflectance in different band combinations have
been widely used to retrieve information on the biophysical properties of vegetation by
using wavelengths of reflected radiation that capitalise on the spectral signatures of
healthy versus diseased vegetation or moisture (Ceccato et al., 2002). Band
combinations such as the Normalised Difference Vegetation Index (NDVI) (Tucker,
1979) and the Normalised Difference Moisture Index (NDMI) have been calculated with
Landsat 8 data to estimate FMC, as they contain bands that are most sensitive to the
water content of plant tissue (Rao et al., 2020; Table 1 and 2). The Visible
Atmospherically Resistant Index (VARI) provides an alternative estimate of greenness
based on visible wavelengths and has been used previously to estimate FMC in
southern California (Schneider et al., 2008) as well as monitor water content (Gitelson
et al., 2002, bands information in Tables 1&2). The NDVI indicator and the normalised
multi-band drought index (NMDI) can be used to estimate soil MC, as those indices
exhibited strong relationships with soil moisture data in previous studies (Wang & Qu,
2007, Tables 1&2). Green Normalized Difference vegetation index (GNDVI) was used in
estimating grassland MC in previous research (Bao et al. 2022), which can be
expanded to estimate FMC in this research (Tables 1&2).
The MODIS (Moderate Resolution Imaging Spectroradiometer) satellite platform
provides daily imagery and has been used previously to monitor FMC. However, its
spatial resolution (250-1000 metres) limits its application in this research because the
planted forest blocks are relatively small compared to other studies (Caccamo et al.,
2011; Yang et al., 2018). Landsat 8/9 and Sentinel 2 have been used to study forest
FMC content in previous research, with their higher spatial resolution data (10 -30
metres resolution), but at the expense of less frequent data acquisition for a given
location (Wang & Qu, 2007; Zhang & Zhou, 2016) (Table 1).

22
Pre- and post-wildfire imagery can be used to distinguish between burned and
unburned forests and can help assess wildfire burn severity and help understand
historical fire behaviours (Key et al., 2006). Due to its 30 m spatial resolution,
approximately 16-day temporal resolution, and extensive library of freely accessible
images dating back to 1984, imagery from the Landsat TM and ETM+ sensors has been
widely used to estimate and map burn severity (Parks et al., 2014). Sentinel 2 imagery
has also been used in burn severity studies (Quintano et al., 2018), and it has higher
spatial (10-30 metres) and temporal resolution (5 days). In the previous research,
differenced Normalised Burn Ratio (dNBR) (Key et al., 2006) and its related form, the
relativized dNBR (RdNBR) (Miller et al., 2007) are the two most used band ratios for
estimating burn severity (Table 2). Instead of using differenced Normalised Burn Ratio
(dNBR), Relativized Burn Ratio (RBR) has been proved as a more appropriate index to
use when comparing burn severity across diverse western US forests, as it uses the
surrounding surviving forest pre- fire conditions as a parameter to adjust the scale of
burn ratio (Parks et al., 2014) (Table 2).
Table 1:Satellite sensors used in this study (Data source: USGS, Sentinel Hub)

23

Table 2: Band indices used in this study (Data source: USGS, Sentinel Hub)
Remote sensing
indices/Application

Satellite

Bands Combination

NDVI: Measure live green
vegetation

Landsat 8/9 (B5-B4)/(B5+B4)

NDMI: Measure crops water
stress level

Landsat 8/9 (B5-B6)/(B5+B6)

NMDI: Measure soil/vegetation
moisture

Landsat 8/9 (B5-(B6-B7))/(B5+B6-B7)

VARI: Estimate vegetation
fraction

Landsat 8/9 (B3-B4)/(B3-B4-B2)

Sentinel 2 (B8-B4)/(B8+B4)

Sentinel 2 (B8-B11)/(B8+B11)

Sentinel 2 (B8a-(B11-B12))/(B8a+(B11-B12))

Sentinel 2 (B3-B4)/(B3-B4-B2)

24

GNDVI: Measure vegetation
greenness

dNBR: Estimate burn severity

Landsat 8/9 (B5-B3)/(B5+B3)
Sentinel 2 (B8-B3)/(B8+B3)
Sentinel 2 Δ(B8-B12)/(B8+B12)
Landsat 8/9 Δ(B5-B7)/(B5+B7)

RBR: Estimate burn severity

Sentinel 2 dNBR/(prefire(B8-B12)/(B8+B12)+1.001)
Landsat 8/9 dNBR/(prefire(B5-B7)/(B5+B7)+1.001)

3. Study Areas
3.1 Prince George and Smithers
The Biogeoclimatic Ecosystem Classification (BEC) zone of the surrounding
forests of Prince George and Smithers is generally classified as the Sub-Boreal Spruce
(SBS) zone (Klinka et al., 1999). The SBS zone is dominated by mature coniferous
forests, including species such as spruce (Picea glauca, Picea engelmannii, Picea
mariana), fir (Abies lasiocarpa), and pine (Pinus contorta) (Natural Resources Canada,
2017). Other tree species found in this zone include larch (Larix laricina), aspen
(Populus tremuloides), and birch (Betula papyrifera) (Natural Resources Canada, 2017).
Cool and moist climates characterise this region, mean annual temperature in the SBS
zone typically ranges from 0°C to 5°C, while mean annual precipitation ranges from 500
mm to 1,500 mm (Ministry of Forests, Lands, Natural Resource Operations and Rural
Development, 2014). However, there can be significant variation within the zone, with
some areas receiving much higher or lower levels of precipitation depending on
elevation, latitude, and local topography (Ministry of Forests, Lands, Natural Resource
Operations and Rural Development, 2014). The terrain is characterised by rolling hills,
flat plateaus and gentle slopes, with well-drained soils and scattered wetlands (Ministry
of Forests, Lands, Natural Resource Operations and Rural Development, 2014).

25
Six research locations (Figures 2 and 3) near Prince George and Smithers in
central BC were selected for this research. The sites were selected for (a) proximity to
Prince George and Smithers; (b) proximity to provincial fire weather stations; and (c)
access to three stand types (open, juvenile, and mature) in close proximity to each other.
The locations selected for this study are approximately 50 km apart in different
directions from Prince George or Smithers, with a provincial fire weather station located
approximately 3.21 – 27.3 km away (Figure 2 and 3). Each research location has three
stand types (Figure 4) that close to each other, with the following characteristics (Burton,
2022):
● Juvenile: Lodgepole pine (Pinus contorta var. latifolia), and interior white spruce
(Picea engelmannii x glauca) have been planted after the primary forest had
been logged. Juvenile forests are dense, with only 2-3 metres separating each
tree; stands are 20–40 years old.
● Open: recently logged, relatively open land with widely space, recently
established pine trees that are less than 5-7 years old.
● Mature: stands that have never been logged; they are characterised by a
preponderance of trees older than 120 years, with various tree species
(lodgepole pine, interior white spruce, subalpine fir and interior Douglas-fir
(Pseudotsuga menziesii var. glauca), with lesser amounts of broadleaf species
such as trembling aspen (Populus tremuloides), black cottonwood (Populus
balsamifera ssp. trichocarpa) and paper birch (Betula papyrifera), at low densities,
and with irregular spatial arrangements.
Each stand’s latitude, longitude, and elevation are shown in Appendix 1. The
name of closest fire weather station to each research location is also provided in
Appendix 2, with its latitude, longitude, elevation, and distance to the research location
presented as well.

26

Figure 2: Prince George field sites (imagery from Google Earth TM)

27

Figure 3: Smithers field sites (Image clipped: Google Earth TM)

28

Figure 4: Tamarac sampling locations near the BCWS Bednesti fire weather station,
showing three stand types (Image clipped: Google EarthTM, RPAS image: Dr.Joseph
Shea)

3.2. Plateau Complex Wildfire, Williams Lake
The Plateau Complex Fire occurred in the Cariboo Regional District of British
Columbia, Canada, approximately 60 kilometres northwest of Williams Lake, and 150
kilometres southwest of Prince George (Figure 5). The fire burned over a distance of
approximately 135 kilometres (84 miles) from south to north, and 70 kilometres (43
miles) from east to west (BC Wildfire Service, 2022). Between 7th July and 1st
September 2017, a few wildfires merged to burn a total of 521,012 hectares covering a

29
broad range of stand types and ages (BC Wildfire Service, 2022). Due to logging and
replanting activities in the area, the wildfire covered a mix of recently harvested open
sites, juvenile replanted sites, and some mature forest blocks (Williams Lake
Community Forest, 2022).
The Plateau Complex Fire provides a suitable case study in which to test
applications of this research, as it covered similar stand types (open, juvenile, and
mature, much of it dominated by lodgepole pine) to the Smithers and Prince George
field sites in central BC. For this case study, the method and results from FMC
observations and remote sensing indices, were used to examine how fire severity
corresponds to areas that exhibited low moisture content/greenness immediately before
the wildfire.
A small subset of the Plateau Complex Fire was selected for this research, as
wildfire conditions can vary substantially across such a large fire perimeter, and smoke
conditions prevented suitable image acquisitions across the entire fire complex. The
subset area included open, juvenile, and mature forests located near each other, similar
to the Prince George and Smithers observation sites, for which suitable pre- and postfire imagery was available for the calculation of fire severity indices. The proximity of
these stands also suggests they would have experienced similar fire weather conditions.

30

Figure 5: Plateau Complex Wildfire map and study area (image captured in Google
EarthTM).
Figures 6 and 7 show pre- and post-fire near-infrared composite images of the
Plateau Complex northwest of Williams Lake, where intense red indicates the vigour of
greenness (juvenile and mature forests), and green/brown indicates the lack of
vegetation (open stands) or burnt areas. The false color composite images in Figures 6
and 7 use the combination of Sentinel-2 bands 8 (near-infrared), 4 (red), and 3 (green).
Note in Figures 6 and 7, that some juvenile forests appeared to act as wildfire refugia,
which remain bright red in both pre-and post-fire images. A comparison of the remote
sensing indices and fuel moisture conditions in the pre-fire area between different stand

31
types can serve as a reference guide for future wildfire vulnerability assessment and
prevention and could help demonstrate the relationship between fire activity and
FMC/remote sensing indices.

Figure 6: False colour composite pre-fire map at Plateau Complex, image captured on
June 11th, 2017, centred on 52.94105 N, -124.01571 W.

32

Figure 7: False colour composite postfire map at Plateau Complex, image captured on
September 4th, 2017, centred on 52.94105 N, -124.01571 W.

4. Methods
4.1 Fuel Moisture Observations
4.1.1 Sample collection
In the summers of 2021 and 2022, samples of foliage (older than 1 year), fine
woody debris, and duff were collected from six research locations near Prince George
and Smithers for this project (Figures 2 and 3). Each sampling location contains three
stand types located near each other. Field samples were taken from open (dominated
new grown pine), juvenile (pine dominated, with some spruce), and mature forests (mix
of pine, spruce, and fir). Higher foliage was collected at heights 4-6 m above the ground
with cutting poles at the juvenile and mature sampling sites to compare with lower

33
foliage. Lower foliage was collected at a height of 1-2 metres. Duff and fine woody
debris were typically picked with a mix of sun-exposed and shaded areas at the Prince
George sites. At the Smithers sites, duff and fine woody debris were collected from
multiple diverse spots in each stand to ensure consistency and obtain a mixture of
moisture data from sun-exposed and shaded areas. Replication of sample trees cannot
be excluded as the sites were chosen randomly at each visit.
In the summer of 2021, at Prince George, twelve samples were collected at each
visit to the juvenile and mature sites: three samples of duff, three of fine woody debris,
and three each for higher (4-6 m) and lower (1-2 m) foliage. Nine samples were
collected for open sites, as there is no higher foliage at open sites. Samples collected at
the start of 2021 were transported in plastic sandwich bags prior to oven drying, and
then were oven-dried in aluminium baking cups. After mid-summer 2021, different sizes
of metal tins with lids were used to collect samples, and these were subsequently stored
and transported in a cooler to reduce moisture loss from the samples, with the same
tins used for sample drying in the oven.
In the summer of 2022, lower-level foliage was not collected due to the increased
sampling frequency and lack of sample tins. Additional characteristics such as tree
species and fuel exposure to the sunlight were documented while collecting the
samples, as they can influence plant water conditions.
Ideally, field sample collection dates would coincide with satellite overpasses to
directly compare field-observed FMC with remote sensing data (Sentinel 2; Landsat 8
and 9). Landsat 8 and 9 (launched in 2022) capture images at 10:00 am local time,
whereas Sentinel 2 collects data at noon local time (USGS, 2022). The sample
collecting period is aimed to be close around noon, as DMC and FFMC are also
calculated at noon (Van Wagner, 1987). However, the collection times often varied due
to weather conditions, transportation time, schedules, and labour constraints. Generally,
samples were collected between 10:30 am and 3:00 pm Pacific Standard Time (PST).

4.1.2 Moisture content measurement
In previous studies, samples of duff, fine woody debris, and foliage were set in an
oven for 48 hours at 60 degrees to dry (Viegas et al., 1992). Before the sample

34
collection began, foliage from the forests at UNBC was collected and tested for two
drying approaches. In the first approach, twelve test samples were measured after
drying for 24 hours at 100 °C and another three days at 100 °C. The MC results from
both approaches were approximately identical, with less than 0.06% moisture change
for foliage and fine woody debris and 0.01% moisture change for duff, compared with
the average dry weight of 5 g for foliage and fine woody debris and 40 g for duff. Due to
a lack of tins and time-saving purposes, samples in this study were dried for 24 hours at
100 °C.
The weight of field samples was measured and recorded in tins after they arrived
at UNBC's Enhanced Forestry Lab (EFL) (Figure 8), and they were sent to dry in an
oven for roughly 24 hours. Tins with dried samples were weighed and recorded on
another day. The weight of the empty tins and lids was subtracted from the dry and wet
weight to get their mass, then Eq.1 was used to calculate FMC. At Prince George, the
scale used to measure woody debris and duff was accurate to 0.01 g. A higher
precision scale (0.001g) was used to measure the weight of the foliage. Samples
collected at Smithers locations were measured with a scale that measured to 0.01 g for
woody debris and duff, and 0.001 g for foliage. Measurements were recorded in field
notebooks and transferred to a spreadsheet along with information on fuel types, tree
species, sunlight exposure, and weather conditions.

Figure 8: Drying process at UNBC EFL lab

35

4.1.3 Moisture content analysis
Observations collected while it was raining, or collected immediately after rainfall
were filtered out as there was water attached to the fuel surface. Foliage from tree
species other than pine, spruce, fir, or Douglas-fir were removed from this analysis as
well. Only juvenile and mature forests were analysed with the remote sensing method
for its correlation with foliage, as their canopy was dense enough to cover the ground.
Foliage data from two summers of fieldwork was separated by species, and their mean,
maximum and minimum values were determined. Using remote sensing to estimate fine
woody debris MC and duff MC was only applied at the open sites, where the forest floor
was fully exposed.
Once two years of data collection were completed, the 3 subsample FMC values
collected for each of foliage, fine woody debris, and duff from each stand at each date
were averaged, then plotted as time series and boxplots to visualize changes through
time and differences between fuel types. Statistical tests (Kruskal-Wallis one way
analysis of variance) were conducted to determine whether the FMC levels varied
significantly among different stands throughout the summers of 2021 and 2022. The
seasonal evolution of FMC in different stands and fuel types may help explain the
existence of juvenile fire refugia. Time-series of FMC values were plotted for each stand
to understand the seasonal evolution of moisture contents in different stand types.
Meanwhile, FMC at different stands and sites was compared with the results from
empirical meteorological models and remote sensing indices, as explained in the
following sections.

4.2 Regional and on-site meteorological data
4.2.1 Fire weather stations
The BC provincial fire weather stations observe several weather parameters
related to fire behaviour. For this research, temperature, relative humidity, wind speed,
and precipitation data were used from fire weather station data. The temporal frequency
of the data collected by these fire weather stations can vary. Generally, the

36
observations are taken at least once per day, usually in the afternoon. However, during
periods of high fire risk or active fire events, the frequency of observations may increase
to multiple times per day or even hourly, depending on the situation. This more frequent
monitoring allows for better tracking of changing fire weather conditions and enhances
the accuracy of fire danger ratings and fire behaviour predictions. Fire weather station
data were obtained from the Pacific Climate Impacts Consortium website
(https://www.pacificclimate.org/data/bc-station-data-disclaimer-0).

4.2.2 In-stand meteorological observations
HOBOTM U23 Pro v2 (Onset Computer Corporation, Bourne, Massachusetts)
temperature and relative humidity sensors (Appendix 3) were installed at all six
sampling locations and in each of the three different stand types. Each HOBO was
mounted on a metal rebar about 30 cm above the forest ground, and their location is
given in Appendix. Full weather stations were also installed at the open sites, measuring
hourly air temperature, relative humidity, precipitation and wind speed at approximately
two metres high (Appendix 4). Open site weather stations had a tipping bucket (except
North Fraser) to collect precipitation data in 2021, though open site weather station
installation occurred after the start of field sampling (July 8th-10th for Smither sites, June
6th – 8th for Prince George sites). Therefore, on-site precipitation and wind speed was
not used to estimate duff and fine woody debris MC in 2021.
For the analysis of 2021 data, in-stand HOBO sensor data (temperature and
relative humidity) were combined with nearby fire weather station data (precipitation
and/or wind speed). In the summer of 2022, tipping buckets (event data only) and wind
speed sensors (recorded every 15 mins) were installed in all stands. Meteorological and
microclimate data will be archived in an open-source repository.

4.2.3 Meteorological data analyses
Daily averages of temperature, relative humidity, and wind speed, as well as
daily accumulated precipitation, were calculated for in-stand and fire weather station
data (Appendix 3).As the data are not normally distributed, Kruskal-Wallis one-way
analysis of variance tests were applied to identify any significant differences between

37
fire weather station and in-stand s weather station data (Section 5.2). As the
meteorological data are not normally distributed, the non-parametric Kruskal-Wallis test
was used to compare means between sites. The null hypothesis is that there is no
significant difference between populations, and a significance threshold of 0.05 was
used to accept (p > 0.05) or reject (p < 0.05) the null hypothesis. Data variance was
also calculated following Eq. 19: (Section 5.2),
(σ²) = Σ[(xᵢ − μ)²] / n

[Eq.19]

where xᵢ represents each individual data point in the dataset, μ represents the mean of
the dataset, and n is the number of observations. Variances are also reported in Sec 5.2.

4.3 FWI empirical model
To test the accuracy of the moisture contents generated by the FWI model in the
study region, four versions of the model used in this research. First, the original FWI
model (Model 1) was used to calculate FFMC and DMC from nearby fire weather station
data to estimate MC with Eq.2 and Eq.17. The second approach (Model 2) used
meteorological data from nearby fire weather stations locally calibrated and standspecific fine woody debris and duff start MC values to estimate MC. The third approach
used partial (Model 3) or full on-site meteorological data (Model 4) from the different
stands and constrained parameters based on local moisture observations to estimate
MC.

4.3.1 Model 1: Fire weather stations and empirical coefficients
The FWI system described in section 3.2 was used with nearest fire weather
station data to estimate the MC of duff and fine woody debris in coniferous stands (Van
Wagner, 1987). This approach is consistent with the current operational use of the FWI
system, and a flow chart outlining this stage of the research is shown in Figure 9. In
2021, FFMC and DMC values from nearby fire weather stations were provided, while in
2022 they were calculated nearby fire weather station data (Figures 2 and Figure 3).
daily expected values of FMC for fine woody debris and duff can be calculated based on
Eq. 2 and 16 with their FFMC and DMC (Viegas et al., 2001, Wotton, 2009).

38
Daily FMC estimates from empirical models, and local/regional meteorological
inputs are compared with the field-based observations of FMC. To compare the
estimated and observed FMC, RMSE and MBE were calculated. The goodness of fit (R2
and p value) for each site and stand type were calculated based on observed MC and
MC predicted from the regression line.

Figure 9: FFMC and DMC application flow chart

4.3.2 Model 2: FFMC and DMC corrected for local moisture contents.
As the FWI model is based primarily on eastern Canada forest types and
climates, previous research has updated parts of the DMC and FMC based on local
conditions (Wotton, 2009). Both DMC and FFMC models require the previous day’s
DMC and FFMC value, to which the drying and wetting effects are added to calculate
the current day’s DMC and FFMC (Eq.2 to Eq.18). To initialise the DMC and FFMC
models at the start of the season, default values of 85 for FFMC and 6 for DMC have
been used in the original model (Van Wagner, 1987). Using Eqs. 2 and 15, this gives
initial moisture contents of 16.3% for fine woody debris and 264.7% for duff. Errors in
the initial MC values used in the FFMC and DMC models would lead to accumulated
errors throughout the summer season.

39
Averaged initial duff and fine woody debris MC at each stand and location
subsamples for 2021 and 2022 are shown in Table 3. In 2021, the average initial MCs
observed in mature stands was 181.7% for duff and 19.2% for fine woody debris, while
for open stands it was 119.1% for duff and 13.3% for fine woody debris. For juvenile
stands, the average initial MC was 137.7% for duff and 26.1% for fine woody debris. In
2022, the average initial MC for mature stands was 281.0% for duff and 61.6% for fine
woody debris, while for open stands it was 193.6% for duff and 28.1% for fine woody
debris. For juvenile stands, the average initial MC was 225.6% for duff and 58.5% for
fine woody debris. While average observed initial values may be similar to those derived
from Eqs. 2 and 15, there is variability between sites and years that may be important
for model estimation.
Table 3: Initial FMC at different sites and stand around Prince George and Smithers in
2021 and 2022, where FWD means fine woody debris.
Duff 2021

FWD 2021

Duff 2022

FWD 2022

Location

Stand

Initial FMC (%) Initial FMC (%)

Initial FMC (%)

Initial FMC (%)

Tamarac

Mature

199.37

17.64

107.6

31.85

North Fraser

Mature

205.23

26.51

281.04

35.7

Stone Creek

Mature

122.47

19.54

246.31

57.78

Chapman

Mature

105.72

13.80

218.65

77.04

McDonnell/Dennis

Mature

261.50

23.20

410.39

63.85

Barren

Mature

196.02

14.25

474.63

103.23

Average

Mature

181.72

19.16

288.77

61.58

Tamarac

Open

79.80

9.73

136.03

16.70

North Fraser

Open

214.29

15.43

185.70

24.49

Stone Creek

Open

104.56

22.06

184.06

9.60

Chapman

Open

152.51

10.13

206.53

30.26

McDonnell/Dennis

Open

62.03

14.21

323.98

14.41

40

Barren

Open

101.30

8.49

125.04

73.19

Average

Open

119.08

13.34

193.56

28.11

Tamarac

Juvenile 97.23

20.36

101.37

16.96

North Fraser

Juvenile 189.06

24.26

227.46

90.88

Stone Creek

Juvenile 191.63

44.96

224.94

83.97

Chapman

Juvenile 66.45

16.12

194.55

62.11

McDonnell /Dennis

Juvenile 128.33

35.59

303.97

40.24

Barren

Juvenile 153.28

15.45

301.38

57.10

Average

Juvenile 137.66

26.12

225.61

58.54

The start averaged duff MC range in central BC was found to vary from 62.0 –
474.6% across different forest stand types, seasons, and weather conditions. As the
DMC was originally developed for pine forests in the eastern US and Canada (Wotton,
2009) and fuel components vary in different regions, Eq. 19 was updated to use the
local in-stand maximum FMC for MCmax and the local in-stand minimum FMC for Et.
By replacing all the parameters “DMC” in the DMC model with Eq.19, duff MC
can be calculated directly from weather conditions. Values of MCmax and E in Eq. 19
were updated based on local measurements of duff MC through summer of 2021 and
2022. Instead of entering default initial DMC to calculate the following, averaged first
observed duff MC from different stands at different locations were entered to calculate
their following MC. With local duff MC values as starting points and updated
maximum/minimum values, it provides an opportunity to monitor the surrounding
different forests' MC based on local conditions.
The previous day’s FFMC is used to estimate to fine woody debris MC with Eq. 2,
then drying and wetting effects are used to calculate the current day’s MC (Van Wagner,
1987). In the FFMC wetting rate, the maximum fine woody debris MC value is set to 250%
(Van Wagner, 1987), whereas it may differ in central BC. For the samples collected
around Prince George and Smithers, the maximum daily averaged fine woody debris

41
MCs at the open stand is 228.4%, 192.0% in the mature stands, and 195.0% in the
juvenile stands, all of which are less than 250%.
Although FFMC does not require minimum fine woody debris MC, like DMC,
FFMC also has a default initial value to start the estimation, which is 16.3%. The initial
averaged fine woody debris MC varies between 8.5 - 103.2% in different stands (Table
3). Resetting maximum MC values in the FFMC model and entering local observation
data as a starting point from Table 3 can provide an opportunity to better estimate fine
woody debris MC locally. These maximum duff MC values, starting DMC, and starting
FFMC as applied to run locally customized FWI indices in each stand type at each
location are presented in Table 6. Subsequent calculations of daily DMC and FFMC
used the standard meteorological data from the nearest fire weather station.
Similar to the analysis of the original FWI model (Section 4.3.1), duff and fine fuel
MC estimated with the updated model parameters were compared with on-site FMC
observations. To evaluate the model skill, R2, p-value, RMSE and MBE were calculated.

4.3.3 Model 3: FWI with local temperature and relative humidity
Model 1 and 2 presented in sections 4.3.1 and 4.3.2 estimate MC with data
measured at a regional fire weather station. However, as fire weather stations are sited
with specific requirements (open areas, measurement heights), and often some
distance away, these cannot be expected to reflect the actual in-stand weather
conditions. On-site relative humidity is higher (particularly at night-time), and wind speed
is lower than observed at the fire weather station. Some of these differences are likely
due to measurement heights: wind speed data were collected at approximately 2.5
meters height for this study, while fire weather station wind speeds are collected at 10
meters height. Total precipitation over the observation period is similar, though
individual event totals vary. Distances between fire weather stations and sampling sites
range between 3.2 and 27.3 km. Hence, the use of regional weather stations potentially
introduces an additional source of error in the estimation of FMC.
In-stand temperature and humidity data were used in Model 3, combined with
nearby fire weather station data for wind speed and precipitation to reduce errors
caused by weather differences. While the FWI formulas were developed to empirically

42
predict in-stand fuel moisture conditions using standardized observations from fire
weather stations, the third empirical model tested here relies partially on-site weather
conditions (temperature and relative humidity), locally updated model parameters and
initial conditions, and is combined with the closest fire weather station data to estimate
in-stand duff and fine woody debris MC.

4.3.4 Model 4: FWI with local weather conditions
Differences in microclimate between juvenile/mature stands and open stands are
also expected, as various forest tree densities and canopy cover lead to different wind
blocking, exposure to sunlight, and interception of rainfall (Karki & Chaudhary, 2018).
Forest stands and their canopy can also influence wind speed and precipitation, which
requires in-stand sensors to capture accurate in-stand data.
In the summer of 2022, wind speed sensors and tipping buckets were installed in
most sites to capture on-site wind speed and precipitation. The new wind speed sensors
and tipping buckets were installed at a height of 2.5 m in all stands and were located
near the 30-cm HOBO sensors (monitoring temperature and relative humidity). In-stand
wind speed data were combined with in-stand temperature and relative humidity to
calculate a localized “in-stand” estimate of fuel moisture contents from FWI indices
(Model 4). RMSE, MBE and goodness of fitting (R2 and p value) from Model 4 were
calculated and compared with other models to check if there is any improvement
statistically.

4.4 Remote sensing and GIS
Multispectral data and remote sensing indices were collected and analysed with
Google Earth Engine (GEE), which provides access to historical satellite images.
Harvested Areas of BC and Vegetation Results Inventory (VRI) data (Government of
British Columbia, accessed May 2023) from the BC data catalogue and Government of
Canada was used in QGIS to select and filter the various forest types based on their
recorded year of logging/ project ages, which can help to determine the forest condition
before wildfires. Although combining Landsat 8 & 9 and Sentinel 2 data will increase the
image capture frequency, there are challenges that prevent the combination of the

43
different platforms (Zhu et al., 2019). Differences between Landsat and Sentinel 2 data,
such as spatial and spectral resolutions, can also limit the usage and introduce errors
when combining them (Zhu et al., 2019).
Bands and remote sensing indices mentioned above from Landsat 8 & 9 and
Sentinel-2 are listed in Tables 1 and 2. Band descriptions, wavelength, and remote
sensing index band combinations are also shown in Tables 1 and 2.

4.4.1 Remote Sensing Indices For FMC
During the first field observation period from late May to September 2021,
Sentinel 2 satellite imagery was collected using Google Earth Engine (GEE). A buffer
with a 30-metre radius was applied to the site of each sampling location, which was
within the range of sample collection, and bands were clipped to the buffered area.
Quality assessment (QA) bands were used to filter out images with clouds and cloud
shadows that could affect the band ratio analysis for Landsat 8 & 9. As Sentinel 2 does
not have a cloud shadow mask, a 2 x 2 km polygon was applied to identify the possible
presence of clouds over the study area, but not directly over the site, as this could
produce cloud shadows. Sentinel 2 scenes that have higher than 10% cloudy pixels in
the 2 x 2 km polygons were considered to potentially contain cloud shadows, and they
were removed from the analysis. The images retained after filtering were exported from
GEE and visually inspected before analysis.
In 2021, the smoke created by wildfires reduced the available satellite images to
use, and field samples were collected at different times than satellite acquisitions.
Therefore, a Locally Weighted Scatterplot Smoothing (lowess) function was used to
generate continuous remote sensing indices values (Moreno et al. 2014) throughout the
summer to match the dates of field sampling. In 2022, sample collection was aimed to
match satellite collection dates. However, the lowess smoothing function was still
applied to avoid mismatches between field sampling data and remote sensing indices
caused by clouds and shadows.
In 2022, Sentinel 2 underwent a shift in its bands and use Harmonized collection
to match the same range as in older scenes (Earth Engine Data Catalog, 2022), which
resulted in its cloud quality assessment (QA) filter not functioning as intended. To

44
address this issue, a simple method of using NDVI to map clouds and cloud shadows
was employed (Tucker, 1979). This method is based on the fact that vegetation reflects
more in the near-infrared range than in the visible range, while clouds and shadows
reflect relatively evenly across all wavelengths (Tucker, 1979; Fonseca & Andrade,
2019; Xiong et al., 2020). Therefore, pixels with low NDVI values are more likely to be
clouds or shadows and can be masked or removed from the analysis. To remove cloudy
pixels from the 2022 Sentinel 2 data, a threshold was first calculated (see below) and
applied to NDVI values calculated for each scene: pixels with NDVI values below this
threshold were classified as cloudy or shadowed.
To determine a NDVI threshold for the detection of cloud and cloud shadows, a
series of clear, cloud/shadow, and cloudy images from the North Fraser site in June
2022 were exported and analysed. NDVI and true colour images were compared in
Appendix 6 to Appendix 8, where A is the true colour image, and B is the NDVI image,
with open (red polygons), old (blue polygon), and juvenile (purple polygon) separated.
When there was no cloud (Appendix 6A), mature and juvenile forests have generally
higher NDVI (0.6-0.8) than open forests (0.4-0.6). When it is cloudy, NDVI is generally
lower than 0.2 (white colour), regardless of the stand types (Appendix 7 and Appendix
8). For cloud shadow, as shown in Figures 12-B, the shadow's NDVI was typically
higher than 0.9 (black colour), which also matched the shadows in the true colour image.
Consequently, when exporting remote sensing indices with 2022 Sentinel 2 data, pixels
with NDVI lower than 0.2 or higher than 0.9 were removed on that image capture day.
Remote sensing indices (NDVI, NDMI, VARI, NMDI, GNDVI, Table 2) were
calculated on an average of the extracted pixel values (partial pixels on edge included)
for buffered areas from GEE (Figure 10). To compare remote sensing indices and
observations of FMC, locally weighted smoothing (lowess) functions were used to
interpolate remote sensing indices between satellite retrieval dates for the 2021 and
2022 summers, as there is a mismatch between sample observation and satellite
visiting days. After applying lowess functions, remote sensing index values from lowess
were paired with the same day averaged observation data at the exact location.
An example of NDMI indices and their lowess smoothers at Tamarac juvenile
stands in 2022 summer is shown in Figure 11. NDMI values from the lowess smoother

45
were compared with sample observation data with a smoothing parameter set to 0.5.
The smoothing parameter controls the balance between the fit to the data and the
degree of smoothing applied to the curve. A lower smoothing parameter corresponds to
a more responsive curve that better fits the data, while higher parameter value
corresponds to a smoother curve (Cleveland, 1974). As there were days with smoke
that couldn’t be removed with QA bands and the NDVI filter, a lowess smoother
parameter set as 0.5 can help to avoid the influence of smoke days.

Figure 10: Tamarac NDVI in different stands in 30 metres pixels in three stand types,
NDVI captured on July 20th, 2022.

46

Figure 11: Example of lowess smoothing of NDMI, at Tamarac juvenile site in 2022
summer, with the smoothing parameter (degree of smoothing) set to 0.25, 0.5, and 0.75.

Remote sensing indices were compared to foliage MC observations at juvenile
and mature sites, and duff and fine woody debris MC values at open sites, representing
exposed/partially exposed ground. To test the performance of the model collaboration,
60% of the data was randomly selected as the training data set to generate linear
regressions, and goodness of fitting (p-value and R2) were calculated. The generated
best-fitting line was then used to calculate the RMSE with the remaining 40% of data.
Randomly selecting training data and calculating the RMSE with the remaining
test data is a common practice in remote sensing studies and has been previously used
in related research (e.g., Zhou et al., 2016). This approach helps verify the results'
reliability and consistency and comprehensively evaluates the remote sensing indices
employed. This comparison can inform the development of empirical functions to
estimate fuel conditions in different stands at a larger scale and help to understand the
fire behaviour from historical wildfires.

47
For foliage in mature and juvenile stands, as well as duff and fine woody debris in
open stands daily-averaged FMC observation data points were compared with index
values from lowess smoothers for the same day; 60% were selected randomly as
training data for the linear regression, their goodness of fit (R² and p-value) was also
calculated, as well as their equations. The remaining 40% observations were used to
test the model and generate error statistics (RMSE).

4.3.2 Mapping Forest types with GIS
Open, juvenile, and mature/old forests have different canopy covers and spectral
characteristics (Cohen et al., 1995). To investigate the relationship between remote
sensing indices and burn severity in stands of different ages, it is necessary to develop
a method for mapping recently logged (open), juvenile, and mature forests prior to the
occurrence of a wildfire. Forest harvest year data was used to map recently logged
forests as it provided more recent and accurate information on logging activity.
Meanwhile, VRI data was used to map juvenile and mature forests as it provided age
information for most forests, including primary forests that have never been logged. For
this study, mature forests were defined as stands older than 60 years, juvenile stands
were logged 10-60 years before 2017, and recently logged areas were logged less than
ten years before 2017. Figure 12 shows that most of the mature, juvenile, and recently
logged forests were successfully mapped for the selected case study region.

48

Figure 12: Stand ages with true colour image on 6th July 2017, centred on 52.94105 N,
-124.01571 W.

4.3.3 Relations between dNBR/RBR and remote sensing indices
As mentioned in section 2.3, dNBR and RBR can both be used to estimate burn
severity after wildfires. For this analysis, QA bands have been used to filter out the
cloudy pixels, and Normalised Difference Water Index (NDWI) has been used to mask
out water bodies such as lakes (Gao, 1996).
To extract the remote sensing indices representing the pre-fire forests/fuel
condition, the image captured at the closest date before the wildfire started should be
chosen. Sentinel 2 data was selected to use when extracting pre-fire indices, as
Sentinel 2 visited the study area (Figure 4) from the Plateau Complex fire area on July
6th, the day before the wildfire started. As wildfires created much smoke, the first cloudfree image from Sentinel 2 after the fire was on September 4th, which was used as a
post-fire image for burn severity calculations. A dNBR value range provided by the
United States Geological Survey (USGS) was used to convert fire severities to burn

49
severity categories (Table 4). In this research, dNBR were classified into five levels,
which are Regrowth (-500 to -101), Unburned (-100 to +99), Low severity (+100 to
+269), Moderate severity (+270 to +650) and High severity (+660). To match the dNBR
range in Table 4, dNBR were scaled by 103.
Table 4: Burn severity levels obtained calculating dNBR, proposed by USGS (Key et al.
2006)
dNBR range (scaled by 103)

Severity level

Enhanced regrowth, high (post fire) -500 to -251
Enhanced regrowth, low (post fire) -250 to -101
Unburned

-100 to +99

Low severity

+100 to +269

Moderate-low severity

+270 to +439

Moderate-high severity

+440 to +659

High severity

+660 to +1300

As RBR was examined and evaluated with 18 different fires in the western USA,
its range classification is different compared to dNBR (Parks et al. 2014). Table 5 shows
the severity level and its averaged RBR range from the results in different fires (Parks et
al., 2014), which was applied in this research as well.
Table 5: Burn severity levels obtained calculating RBR, proposed by Parks et al., 2014.
Severity level
Unburned
Low severity
Moderate severity
High severity

RBR range
<35
35-130
130-298
>298

Remote sensing indices such as NDMI and NDVI were extracted from the prefire image to compare fire severity among the different stands. As this method produces
a large quantity of data points, hexbin plots of indices versus burn severity were plotted
to show data density, frequency distribution (histograms at x and y axis), and relations

50
between the two elements. Hexbin plots are commonly used to reduce noise in the data
by showing the overall distribution and patterns in the data, rather than individual data
points (Camacho, 2014). Linear or non-linear fits were not attempted in the analysis of
estimated moisture content and fire severity, and this is discussed in Section 6.

5. Results
5.1 Fuel Moisture Contents
Summaries of the daily average MC for different observations sites, forest stands
(juvenile, mature, and open), different types of fuel (duff and fine woody debris), as well
as statistical results are presented in Table 6. Information on the total number of daily
averaged observations for each category is also included in Table 9, along with the
maximum and minimum FMC values. The maximum FMC for duff was 439.27 %, which
was found in mature stand; the minimum MC for duff was 14.38 %, and it was in a
mature stand (Table 6).
The daily averaged MC for duff was also listed in Table 6, and it was not normal
distributed in all stands. At mature stand had 163.76 % MC compared with juvenile
(143.56 %), and the difference was statistically significant (p < 0.05) with Kruskal-Wallis
test. The daily averaged FMC for duff at open stands is 14.03 % lower compared with
juvenile, and the difference was statistically significant (p < 0.05).
Information on the daily averaged FMC for fine woody debris was also in Table 6.
The average fine woody debris MC for juvenile stands was 50.19 %, while for mature
stands it was 1.17 % lower, which was not statistically significant (p > 0.05). However,
for open stands, the daily averaged MC for fine woody debris was 20.11 % lower
compared to juvenile stands, and this difference was statistically significant (p < 0.05).
The maximum MC for daily averaged fine woody debris was 228.38 %, which was found
in an open stand, and the minimum FMC was 2.17 %, also in an open stand.

51
Table 6: The maximum, minimum of daily averaged MC, variance, and averaged of
daily averaged MC for duff and woody debris from different stands in 2021 and 2022.
Forest Stand

Juvenile Mature

Open

Juvenile

Mature

Open

Fuel Type

Duff

Duff

Duff

Fine woody debris Fine woody debris Fine woody debris

Total Number

132

131

132

132

132

132

Maximum MC (%) 407.44

423.68

439.27

194.98

192.04

228.38

Minimum MC (%) 28.75

29.87

14.38

8.64

9.53

2.17

Averaged MC (%) 143.56

163.76

129.57

50.19

51.9

49.02

1428.62

1300.20

Variance (%2)

6918.07 8082.44 7514.89 1515.28

As different types of foliage species were collected during the summers of 2021
and 2022, daily average foliage MC, as well as maximum and minimum MC values for
each stand and species type are shown in Table 7. Kruskal-Wallis tests were conducted
to determine if there were any significant differences in daily averaged FMC among the
different types of foliage species, as none of the dataset is normal distributed.
At the juvenile stand, where pine was the dominant species type, its daily
averaged MC was compared with that of spruce and fir. Pine foliage had an average
MC of 112.38%, which is 0.08% lower than fir and 8.45% higher than spruce (Table 7).
However, the difference was not statistically significant, as both p-values were higher
than 0.05. Fir was used as the reference species to compare with other species types at
the mature stand (Table 7), and no statistically significant differences were observed,
with all p-values higher than 0.05. At the open stand, pine was compared with spruce
and fir (Table 7). Similar to the juvenile and mature stands, the p-values were both
higher than 0.05, indicating no statistically significant difference in foliar among MC
species.
Table 7: The maximum, minimum of daily averaged MC, and averaged of daily
averaged MC for foliage from different stands around Smithers and Prince George in
2021 and 2022, where DFir means Douglas-fir.
Foliage
type
Forest

Pine

Spruce

Fir

Pine

Spruce

Fir

DFir

Pine

Spruce

Fir

Juvenile

Juvenile

Juvenile

Mature

Mature

Mature

Mature

Open

Open

Open

52
stand
Total
Number
Maximum
MC (%)

111

21

23

6

49

108

6

113

3

15

151.40

156.81

195.82

112.27

237.12

191.65

299.26

233.85

226.26

161.32

Minimum
MC (%)

82.63

71.49

73.57

101.39

50.28

74.02

101.27

41.4

105.60

91.32

Averaged
MC (%)

112.38

103.93

112.47

105.65

108.29

111.34

143.91

124.78

148.71

115.67

Variance
(%)2

281.83

611.01

1184.78

22.79

806.39

1821.17

5902.24

916.48

2662.78

418.99

As there is no statistically significant difference in MC between foliage species
types in all stands, different foliage species types were combined to check their FMC
difference in different stands. The daily averaged maximum and minimum MC of foliage
in different stands in all locations, as well as their observation number, are presented in
Table 8. Open stands have the highest maximum MC (233.9%) and minimum MC
(41.4%). When comparing the averaged MC of foliage with their daily averaged data,
mature forests have -2.85% MC lower than juvenile stands (112.34%), however, the
difference is not statistically significant with (p > 0.05). Open stands have 12.62% MC
higher than juvenile forests, and the difference is statistically significant, with p < 0.05.
Table 8: The maximum, minimum of daily averaged MC, and averaged of daily
averaged MC for foliage from different stands in 2021 and 2022.
Forest Stand
Total Number
Maximum MC (%)
Minimum MC (%)
Averaged MC (%)
Variance (%2)

Juvenile Mature Open
115
116
120
152.68 194.83
233.85
74.88
75.71
41.40
112.34 109.49
124.96
694.40 1066.87 1197.82

The time series of daily averaged FMC (consisting of three individual
measurements averaged in each stand and study location) observed in duff, woody
debris, and foliage for the summers of 2021 and 2022 are presented in Appendix 9 and
Figure 13. In both 2021 and 2022, daily averaged MC in the duff layer is significant
higher in mature forest stands, followed by juvenile and open stands. Fine woody debris
at the mature and juvenile sites have a similar MC, with no statistically significant

53
difference (p > 0.05). However, daily averaged fine woody debris MC at open stands are
lower than mature and juvenile stands, with statistically significant difference (p < 0.05).

Figure 13: Daily average FMC in 2022 for different stands and fuel types at the Prince
George and Smithers study locations. The number of observations averaged to create
each point varies between 3 and 15, depending on the number of sites visited.
When comparing different fuel types and stands at each location, similar
statistical results were found in Figure 14 to 16. To compare the FMC difference in
different stands at each location, as the data is not normal distributed, Kruskal-Wallis
test were also applied.
For foliage MC, open stands have higher MC compared to juvenile stands, and
the difference is statistically significant at Barren, Tamarac, Chapman, and McDonnell

54
sites (p < 0.05). The difference between juvenile and mature stands is generally not
statistically significant, except at Chapman, where the mature stands have significantly
higher MC (p < 0.05), and at McDonnell/Stone Creek, where the juvenile stands have
significantly lower MC (p < 0.05).
For duff MC at different stands at each location, juvenile stands have significantly
higher MC than open stands at most locations (Barren, Dennis, North Fraser, and Stone
Creek, with p < 0.05), significantly lower MC at Chapman and McDonnell, and no
statistically significant difference at Tamarac (p > 0.05). When comparing juvenile and
open stands, most locations show no significant difference (Tamarac, Dennis, North
Fraser, and Stone Creek with p > 0.05). However, at Barren and Dennis, juvenile stands
have significantly higher averaged MC than open stands (p < 0.05), while at Chapman,
they have significantly lower averaged MC than open stands (p < 0.05).
When comparing fine woody debris, juvenile stands have higher daily averaged
MC than open stands at each location, and the difference is also statistically significant
(p < 0.05). However, the difference is only significant at Barren and North Fraser sites
when comparing mature and juvenile stands.
In summary, this study finds that location and stand differences play a crucial role
in daily averaged FMC values for different fuel types. In the next sections, daily FMC
values at each fuel types, location, and stand types were tested and compared with
different methods.

55

Figure 14: Boxplots of daily averaged duff MC at different stands and locations in 2021
and 2022 around Prince George and Smithers. Lower box limits (first quartile -1.5 * IQR
(Interquartile range), upper box limits (third quartile +1.5 * IQR), outliers (circles), mean
values (green triangles), and median values (yellow horizontal lines) are shown.

56

Figure 15: Boxplots of woody debris MC at different stands and sample collection sites
in 2021 and 2022 around Prince George and Smithers.

57

Figure 16: Boxplots of foliage MC at different stands and sample collection sites in 2021
and 2022 around Prince George and Smithers.

58

5.2 Meteorological Data
As the distance between fire weather stations and sample collecting sites ranges
between XX and XX km, weather conditions may vary between fire weather stations
and in-stand locations. Figure 17 shown an example of daily weather conditions plots
between the open weather station at Barren site and the Houston fire weather station in
the summer of 2022. The difference of relative humidity and precipitation between
Barren open site and Houston fire weather station is not significant (p>0.05), however,
there is a significant difference (p<0.05) for temperature and relative humidity.

59

Figure 17: Weather condition comparison between the BC fire weather station (FWS,
Houston) and onsite weather station in 2022 at Barren, BC.

Figure 18 demonstrates the differences in daily weather conditions at the Barren
site in 2022, and summaries of the measured meteorological data are given in Appendix
4.

60
Comparisons of measured meteorological data from the six different locations
and various stands to determine if significant differences exist between stands
(Appendix 4).
Mature, juvenile, and open stands had average temperatures of 13.95°C,
14.17°C, and 15.29°C. The mean temperatures were not significantly different between
mature and juvenile stands (p>0.05, the null hypothesis cannot be rejected), but the
open stand had significantly higher temperatures.
The mature stand had an average relative humidity of 79.64%, while the juvenile
stand had a significantly lower average of 72.71%. In open stands, the average relative
humidity was 72.62%. The difference between open stands and both the mature and
juvenile stands was also statistically significant (p < 0.05).
The daily averaged wind speed for the six locations in 2022 was also analyzed
using the Kruskal-Wallis test due to the non-normal distribution of the data. The mature
stand had an average wind speed of 0.10 m/s. In the juvenile stands, the average wind
speed was 0.03 m/s. The difference between the mature and juvenile stands was not
statistically significant (p > 0.05). However, the open stand had an average wind speed
of 0.79 m/s, and the difference between the open stand and the juvenile stands was
found to be significant (p < 0.05).
Finally, the daily accumulated precipitation was compared among the stands
using the Kruskal-Wallis test. The mature stand had an average daily accumulated
precipitation of 1.27 mm. The juvenile stands had an average of 1.54 mm of
accumulated precipitation. The difference between the mature and juvenile stands was
statistically significant (p < 0.05). In the open stands, the daily accumulated precipitation
was 1.57 mm, slightly higher than the juvenile stands, but the difference was not
statistically significant (p > 0.05).

61

Figure 18: Weather condition comparison between on-site weather stations at different
sites in 2022 at Barren, BC.

62

5.3 Estimates of FMC from FWI
5.3.1 Model 1: Original model with fire weather stations and empirical
coefficients
Based on the methods in section 4.2, FMC for fine woody debris and duff were
estimated from the fire weather station using the coefficients described in the empirical
DMC and FFMC models. Modelled and observed FMC were compared for the different
stands and years (Figure 19) using goodness-of-fit statistics. In 2021, the estimated duff
MC was generally lower than the observed MC, with a large discrepancy between the
best-fitting and one-to-one lines. However, in 2022, the estimated duff MC was
generally like the observed MC, with plots closely surrounding the one-to-one line. The
estimated and observed values for fine woody debris were generally close. However, in
2022 estimate fine woody debris MC in juvenile and mature stands was lower than
observed in general.
Model performance (Table 9) varies among different stand types (open, mature,
juvenile) and fuel types (duff and woody debris). Mean Bias Error (MBE) values for duff
and woody debris range from -58.24 to 31.87 %, with the lowest MBE of -0.95 %
observed in the open stand type with woody debris in 2021. RMSE values range from
26.78 to 85.35 %, with the lowest RMSE of 26.78 % observed in the juvenile stands with
woody debris in 2021. The R² values for the duff multiple regression tests of observed
and predicted values range from 0.17 to 0.44, depending on the stand type and year,
with highest R² observed in the open stands in 2021. For fine woody debris, the R²
values for multiple regression tests of observed and predicted values range from 0.17 to
0.65, with highest R² observed in the open stands in 2021. All p-values are less than
0.001, which suggests that the relationships between the predictor variable and the
outcome variable are statistically significant for all combinations of stand and fuel type.
It can be observed that the ranges of RMSE and MBE are substantial for duff and
fine woody debris. However, fine woody debris exhibits better statistical agreement than
duff, particularly in open stands. The agreement between estimated and observed duff
FMC improved in 2022 compared to 2021, while the statistical agreement for fine woody

63
debris in 2021 was superior to that in 2022. Across years and fuel types, the RMSE and
MBE in open sites are generally better than those in juvenile and mature sites.

Figure 19: Estimated vs observed FMC from as-is weather empirical model at different
stands, where black line indicates the one-to-one lines, orange line indicates the best
fitting line.

64
Table 9: Errors in estimated MC for different stand and fuel types for the ‘as-is’ empirical
model (FFMC/DMC) in 2021 and 2022. Mean Bias Error (MBE), Root Mean Squared
Error (RMSE), R², p, and sample size (N) are given. “FWD” means fine woody debris, “x”
is estimated FMC and “y” is observed FMC.
Model 1

Year

Stand Type Fuel Type MBE
(%)

RMSE
(%)

R²

p

Equation

N

2021

Open

duff

-56.22 83.94

0.44

<0.001

y = 0.71*x – 14.7

94

2021

Mature

duff

-39.25 78.08

0.22

<0.001

y = 0.41* x +49.08

93

2021

Juvenile

duff

-58.24 85.35

0.17

<0.001

y = 0.36*x + 41.63

94

2022

Open

duff

-30.82 68.93

0.42

<0.001

y = 1.02*x -40.79

106

2022

Mature

duff

17.16

82.99

0.35

<0.001

y = 1.07*x + 2.59

106

2022

Juvenile

duff

-6.49

59.95

0.39

<0.001

y = 0.99*x – 4.03

106

2021

Open

FWD

-0.95

27.72

0.32

<0.001

y = 0.56*x + 12.28

93

2021

Mature

FWD

13.77

30.26

0.48

<0.001

y = 0.81*x + 19.17

93

2021

Juvenile

FWD

15.37

26.78

0.65

<0.001

y = 0.98*x + 15.68

93

2022

Open

FWD

8.94

36.20

0.17

<0.001

y = 0.64*x + 12.13

128

2022

Mature

FWD

31.13

44.90

0.31

<0.001

y = 0.90*x + 27.59

128

2022

Juvenile

FWD

31.87

47.42

0.34

<0.001

y = 0.97*x + 27.8

128

5.3.2 Model 2: FFMC and DMC correction for local MC
Errors may increase when using the empirical weather model based on boreal
and eastern Canada to estimate FMC in central BC. To mitigate this issue, the default
maximum and minimum duff and fine woody debris MC (Table 6) was modified in the
FFMC and DMC equations to minimum and maximum daily averaged values from all
locations in 2021 and 2022 to reduce errors caused by varying fuel conditions across

65
different stands. Furthermore, instead of using a default starting value of FFMC and
DMC, the average MC of duff and fine woody debris on the first day of data collection
was used to estimate the following day's MC in both 2021 and 2022 (Figure 20).
However, the estimated MC for fine woody debris in juvenile and mature stands in 2022
is smaller than the levels observed.

Figure 20: Estimated versus observed FMC from FWI (original model corrected by local
MC), where black line indicates the one-to-one lines, orange line indicates the best
fitting line.
MBE for estimated MC in fine woody debris and duff in 2021 and 2022 range
from -68.19% to 27.75%, with the lowest MBE value (-0.17) occurring in the open stand

66
type with woody debris in 2022. RMSE values range from 27.22 to 94.33%, with the
lowest (27.20%) occurring in the juvenile stand type with woody debris in 2022. In terms
of R², duff has highest R² (0.60) at mature stands in 2022, while fine woody debris has
highest R² (0.65) at juvenile stands in 2021. All p-values presented in the Table 10 are
less than 0.001, indicating the relationships between the predictor variable and the
outcome variable are statistically significant for all combinations of stand and fuel type.
With Model 2 (local MC correction), fine woody debris MC generally exhibits
better statistical agreement with predicted values across all stands and years than duff.
In 2021-2022, the estimated errors for fine woody debris MC in Table 10 are similar or
slightly lower than those in Table 9, whereas the MBE and RMSE for duff in Table 10
were slightly greater than in Table 9.
An improvement for fine woody debris MC prediction demonstrates the potential
for more accurate results using local MC observations. However, large values for MBE
and RMSE indicate that there is still room for improvement.
Table 10: Errors in estimated MC for different stand and fuel types for weather empirical
model (FFMC/DMC) at different stands in 2021 and 2022, corrected by local MC. Mean
Bias Error (MBE), Root Mean Squared Error (RMSE), R², p, and sample size (N) are
given. “FWD” means fine woody debris, “x” is estimated FMC and “y” is observed FMC.
Model 2
Year

Stand Type Fuel Type

MBE

RMSE

R²

p

Equation

N

2021

Open

duff

-38.01

78.96

0.43

<0.001

y = 0.6*x +21.35

93

2021

Mature

duff

-45.89

84.80

0.32

<0.001

y = 0.41*x +47.4

94

2021

Juvenile

duff

-48.47

87.51

0.25

<0.001

y = 0.36*x + 44.44

94

2022

Open

duff

-36.82

94.33

0.45

<0.001

y = 0.74*x – 13.90

106

2022

Mature

duff

-68.19

72.20

0.60

<0.001

y = 0.88*x – 7.45

106

2022

Juvenile

duff

-45.12

75.68

0.51

<0.001

y = 0.83*x – 7.70

106

2021

Open

FWD

-0.17

27.20

0.33

<0.001

y = 0.59*x + 11.81

94

67
2021

Mature

FWD

15.97

30.20

0.48

<0.001

y = 0.97*x + 16.74

93

2021

Juvenile

FWD

16.97

36.13

0.65

<0.001

y =1.12*x + 12.49

93

2022

Open

FWD

24.99

41.87

0.19

<0.001

y = 0.69*x + 10.93

106

2022

Mature

FWD

2.26

43.85

0.33

<0.001

y = 1.00*x + 25.00

106

2022

Juvenile

FWD

37.75

47.42

0.34

<0.001

y = 1.07*x + 25.68

106

5.3.3 Model 3: FFMC and DMC with local temperature and relative humidity
The previous section found that the use of FFMC and DMC models resulted in
errors due to differences in weather conditions between the closest fire weather station
and the sample observation sites. To address this issue, local weather data
(temperature, relative humidity, precipitation, and wind speed) from sensors or weather
stations were incorporated to determine if this could improve the accuracy of estimating
FMC. Meanwhile, the use of initial MC of duff and woody debris, as well as FMC limits
applied in Model 2 was continued in the model with local weather conditions.
In 2021 and 2022, HOBO sensors were installed at all sample collection sites
and recorded temperature and relative humidity every 15 minutes at 30 cm above the
ground. To estimate FMC, wind speed and precipitation data from the nearest fire
weather station were combined with the in-stand HOBO sensors' temperature and
relative humidity data, as not all sites had installed wind and precipitation sensors in
2021.
The estimated and observed FMC values in 2021 and 2022 are compared in
Figure 21. The black line represents the most accurate estimation, and the orange line
represents the best-fitting line. Unlike Figure 20, data from the in-stand HOBO sensors
was used to replace the temperature and relative humidity inputs into the DMC and
FFMC models.

68

Figure 21: Estimated versus observed FMC with local in-stand temperature and relative
humidity, where orange line indicates the best fitting line.
Table 11 presents MBE, RMSE, R², p values and their linear regression
equations for Model 3 in 2021 and 2022. The MBE ranges from -130.23 to 47.31%
(Table 11), with the lowest MBE at -9.13% when estimating duff MC at mature forests in
2022. The RMSE values range from 59.95 to 158.09%, where the lowest error is for
estimating duff MC at mature stands in 2022 as well (Table 11). For fine woody debris,
MBE ranges from -18.45 to 26.63%, with lowest MBE at 1.54 at open stands in 2021. In
terms of RMSE, it ranges from 30.20 to 44.93%, with lowest RMSE at open stand in
2021 when estimating fine woody debris MC.

69
The R² ranges for duff and fine woody debris are 0.08-0.72 and 0.08-0.69,
respectively. The p-values for most the stands are less than 0.001, indicating that the
relationships between the predictor and response variables are statistically significant,
except estimating duff MC in 2021 at mature juvenile stands and fine woody debris in
2022 at all stands. The highest R² for estimating duff MC is the mature stand type in
2022 with an R² of 0.72. For fine woody debris, the highest R² is at mature stand type in
2021 with an R² of 0.69.
After updating the local temperature and relative humidity data, the statistical
results for some stands improved, such as estimating duff MC at mature stands in 2022,
indicating that the method improved accuracy in some aspects. However, due to
significant differences in wind and precipitation between open and juvenile/mature
stands, combining the local in-stand data with data from nearby fire weather stations
could result in errors.
Table 11: Errors in estimated MC for different stand and fuel types for the weather
empirical model (FFMC/DMC) calibrated with local temperature and relative humidity,
as well as local fuel correction. Mean Bias Error (MBE), Root Mean Squared Error
(RMSE), and sample size (N) are given. “FWD” means fine woody debris, “x” is
estimated FMC and “y” is observed FMC.
Model 3
Year

Stand Type Fuel Type

MBE

RMSE

R²

p

Equation

N

2021

Open

duff

-44.84

86.99

0.31

<0.001

y = 0.52*x + 20.7

71

2021

Mature

duff

-127.86

158.09

0.18

<0.001

y = 0.23*x + 48.05

60

2021

Juvenile

duff

-130.23

153.63

0.08

0.021

y = 0.14*x + 62.56

64

2022

Open

duff

47.31

71.59

0.38

<0.001

y = 0.75*x + 65.82

62

2022

Mature

duff

-9.13

59.95

0.72

<0.001

y = 0.64*x + 58.85

64

2022

Juvenile

duff

11.44

60.30

0.54

<0.001

y = 0.66*x + 68.34

56

2021

Open

FWD

-1.54

30.20

0.30

<0.001

y = 0.52*x + 14.14

70

70
2021

Mature

FWD

-18.45

35.46

0.69

<0.001

y = 0.54*x + 10.25

61

2021

Juvenile

FWD

-13.23

34.07

0.65

<0.001

y = 0.54*x + 14.68

64

2022

Open

FWD

9.39

36.59

0.11

0.009

y = 0.62*x + 16.14

62

2022

Mature

FWD

23.09

44.93

0.13

0.003

y = 0.63*x + 33.92

64

2022

Juvenile

FWD

26.63

44.87

0.08

0.033

y = 0.38*x + 43.99

56

5.3.4: Model 4: FFMC and DMC with local weather conditions
In 2022, all sample collection locations were equipped with tipping bucket rain
gauges and wind speed sensors, in addition to HOBO sensors that recorded
temperature and relative humidity. The data collected from these devices were utilised
to calculate the MC of duff and fine woody debris. Integrating data from multiple sources
allowed for a more comprehensive assessment of fuel conditions, as it considered
precipitation, wind speed, temperature, and relative humidity (Figure 22). In juvenile and
mature stands, tipping buckets were only installed during in the middle of the sample
collection season, and sample numbers were reduced in Model 4 compared to the other
three models.
When comparing the RMSE and MBE values in Model 4 (Table 12) to those of
Models 1, 2, and 3, it can be seen that for both duff and fine woody debris fuel types,
the agreement between predicted and observed values are generally better. For duff,
the smallest RMSE value in Table 15 was 66.11% for duff in mature stands. For fine
woody debris, the lowest RMSE is 21.13 when estimating woody debris MC at juvenile
stands. In terms of the of MBE (Table 12), the range was from -60.71 to -35.84 (duff),
with a lowest MBE at mature stands. The MBE range for fine woody debris was lower (1.32 to 6.20%) than for duff, where the lowest values (best agreement) as found for
open stands (-1.32%). The highest R² was the mature stand with an R² of 0.71 when
estimating duff MC. For fine woody debris, the highest R² was the juvenile stand type
with an R² of 0.76. All the stands have statistically significant relationships between the
predictor and response variables of the linear regression at the p-value limit of 0.001.

71
This suggests that Model 4 produces better predictions regarding the spread of
residuals (errors) compared to Models 1, 2, and 3.

Figure 22: Estimate vs observation FMC with local weather conditions, where black line
indicates the one-to-one lines, orange line indicates the best fitting line.
Table 12: Errors in estimated MC for different stand and fuel types for weather empirical
model (FFMC/DMC) with local temperature, relative humidity, wind speed, and
accumulated precipitation, as well as local fuel correction. Mean Bias Error (MBE), Root
Mean Squared Error (RMSE), and sample size (N) are given. “FWD” means fine woody
debris, “x” is estimated FMC and “y” is observed FMC.
Model 4
Year

Stand Type

Fuel Type

MBE

RMSE

R²

p

Equation

N

2022

Open

duff

-60.71

89.40

0.41

<0.001

y = 0.53*x + 25.67

71

2022

Mature

duff

-35.84

66.11

0.71

<0.001

y = 0.71*x + 24.54

60

2022

Juvenile

duff

-56.23

85.53

0.40

<0.001

y = 0.60*x + 22.40

64

2022

Open

FWD

-1.32

27.20

0.43

<0.001

y = 0.75*x + 5.92

70

2022

Mature

FWD

6.20

25.26

0.61

<0.001

y = 0.84*x + 12.09

61

2022

Juvenile

FWD

1.65

21.13

0.76

<0.001

y = 0.82*x + 9.99

64

72

5.4 Estimates of FMC from Remote Sensing
5.4.1 Estimating FMC with Sentinel 2 data in 2021
In 2021, Sentinel 2 was used as the source of remote sensing data compared
with field observation data, while in 2022, both Landsat 8 & 9 and Sentinel 2 data were
used (Table 13). When extracting remote sensing indices from GEE, the time series
was set between May 1st to September 30th in 2021 and May 1st to September 3rd in
2022 to cover and match the sample collection period. Due to the smoke created by the
wildfires, valid images available in 2021 (11.48 - 32.79 %) were fewer than in 2022 (50 65.38 %) when using Sentinel 2 data (Table 13). Landsat 8 & 9 data have fewer total
image numbers (31 - 62 images), and the valid percent was generally low (12.77% 17.65%).
Table 13: Study area’s cloud/shadow free images in the summer of 2021 and 2022 from
Smithers and Prince George
Year

Satellite

Site name

Valid

Total

Valid Percent (%)

2021

Sentinel 2

Tamarac

15

62

24.19

2021

Sentinel 2

North Fraser

18

61

29.51

2021

Sentinel 2

Stone Creek

20

61

32.79

2021

Sentinel 2

Chapman

8

61

13.11

2021

Sentinel 2

McDonnell

7

61

11.48

2021

Sentinel 2

Houston

8

61

13.11

2022

Sentinel 2

Tamarac

27

53

50.94

2022

Sentinel 2

North Fraser

31

50

62.00

2022

Sentinel 2

Stone Creek

26

51

50.98

2022

Sentinel 2

Chapman

33

52

63.46

73

2022

Sentinel 2

Dennis

27

54

50.00

2022

Sentinel 2

Houston

34

52

65.38

2022

Landsat 8&9

Tamarac

6

62

9.68

2022

Landsat 8&9

North Fraser

5

31

16.13

2022

Landsat 8&9

Stone Creek

6

47

12.77

2022

Landsat 8&9

Chapman

6

34

17.65

2022

Landsat 8&9

Dennis

5

32

15.63

2022

Landsat 8&9

Houston

5

32

15.63

Example maps for the five indices used in this research are shown in Figure 23.
For NDMI and NMDI, juvenile and mature forests generally have higher index values
than open stands, while VARI and GNDVI have lower index values instead (Figure 23).
However, the situation might change from other days’ satellite imagery and another
location.

74

A

B

C

D

E

Figure 23: Example of remote sensing indices. A) NMDI, B) VARI, C) GNDVI, D) NDMI,
E) NDVI at North Fraser sites on June 2nd , 2022, where red polygons indicate the open
stand, purple polygons indicate juvenile stands, and blue polygons indicate mature
stands. Darker green colours indicate higher indices values (denser and healthy forest,
opposite with VARI), and whiter colour indicating lower indices values (bare vegetation
cover, opposite with VARI).
Figures 24 - 27, blue points are the training data, and the green line is their best
fitting line, while orange points are the test points.

75

Figure 24: Foliage MC compared with five remote sensing indices at juvenile stands
with Sentinel 2 data in 2021, where blue points are train data, green lines are the linear
regression based on blue points, and orange points are test data.

Figure 25: Foliage MC compared with five remote sensing indices at mature stands with
Sentinel 2 data in 2021, where blue points are train data, green lines are the linear
regression based on blue points, and orange points are test data.

76

Figure 26: Duff MC compared with five remote sensing indices at open stands with
Sentinel 2 data in 2021, where blue points are train data, green lines are the linear
regression based on blue points, and orange points are test data.

Figure 27: Fine woody debris MC compared with five remote sensing indices at open
stands with Sentinel 2 data in 2021, where blue points are train data, green lines are the
linear regression based on blue points, and orange points are test data.

Regression models for estimating MC from remote sensing indices using the
2021 data are mostly poor (Table 14). Only one relationship (NMDI versus foliage MC
in juvenile stand) appears to be statistically significant, with (R² = 0.334 and p < 0.05).

77
Other than NDMI for juvenile stands, all the other indices R² values are less than 0.10,
and/or p-values are greater than 0.05. In general, RSME calculated from independent
observations and the regression model are lower in juvenile stands than in mature
stands. VARI-predicted foliage MC in juvenile stands had the lowest RMSE (12.76),
while GNDVI at the mature site had the lowest RMSE at 14.61 (Table 14).
Table 14: Statistical results of remote sensing indices with second year foliage MC at
juvenile and mature stands (2021 Sentinel 2), where “y” is the observed foliage MC.
Significant relations (p < 0.05) shown in bold.
Index

Stand Type

R²

p

RMSE

Equation (train)

N train

N test

NDVI

Juvenile

0.009

0.2198

13.58

y = -32.23*NDVI + 134.96

27

19

NDMI

Juvenile

0.128

0.0671

13.97

y = 64.82*NDMI + 80.73

27

19

NMDI

Juvenile

0.334

0.0016

17.86

y = 141.40*NMDI + 13.53

27

19

VARI

Juvenile

0.087

0.0500

12.76

y = 55.05*VARI + 138.83

27

19

GNDVI

Juvenile

0.040

0.3170

18.88

y = 76.34*GNDVI + 67.51

27

19

NDVI

Mature

0.007

0.6720

23.50

y = -72.51*NDVI + 161.45

27

19

NDMI

Mature

0.000

0.9710

15.92

y = 18.95*NDMI + 112.71

27

19

NMDI

Mature

0.005

0.7170

43.00

y = 66.62*NMDI + 70.22

27

19

VARI

Mature

0.013

0.3178

20.35

y = -116.98*VARI + 63.75

27

19

GNDVI

Mature

0.012

0.5930

14.61

y = -87.54*GNDVI + 157.48

27

19

Statistical results for duff and woody debris MC models in open stands are
shown in Table 15. No significant relations (p < 0.05) could be identified. At open sites,
NMDI had a stronger correlation (R² at 0.046, p at 0.22) with fine woody debris. In
comparison, GNDVI had a slightly stronger correlation when estimating duff MC than
other indices, as indicated by their relatively higher R² value (0.092) and lower p-value

78
(0.07). Duff predictions had generally much higher RMSE (around 41.64 – 63.29) than
for fine woody debris (20.90 - 36.43) with all indices. However, all indices in Table 15
were not good enough to estimate duff and woody debris MC, with their lower R² (less
than 0.1) and higher p-values.
Table 15: Statistical results of remote sensing indices with fine woody debris/ duff MC at
open stands (2021 Sentinel 2), where FWD means fine woody debris, y is observed
FMC. No significant relations were identified.
Index Fuel Type

R²

p

RMSE

Equation (train)

N train

N test

NDVI

duff

0.002

0.0800

63.29

y = 35.22*NDVI +43.78

36

24

NDMI

duff

0.020

0.6803

41.64

y = 95.98*NDMI +59.00

36

24

NMDI

duff

0.016

0.5537

55.82

y = 164.38*NMDI – 9.27

36

24

VARI

duff

0.009

0.3133

57.84

y = 36.71*VARI + 79.70

36

24

GNDVI

duff

0.092

0.0728

50.83

y = 402.45*GNDVI-132.72

36

24

NDVI

FWD

0.002

0.0661

34.86

y = 11.66*NDVI + 14.76

35

24

NDMI

FWD

0.020

0.6574

20.90

y = 48.33*NDMI + 19.52

35

24

NMDI

FWD

0.046

0.2150

21.68

y = 145.27*NMDI - 46.97

35

24

VARI

FWD

0.005

0.1599

36.43

y = 10.16*VARI + 23.41

35

24

GNDVI

FWD

0.014

0.4761

32.34

y = 55.75*GNDVI – 4.73

35

24

5.4.2 Estimating FMC with Landsat 8 & 9 data in 2022.
With the launch of Landsat 9 in 2022, image frequency increased, and indices
could be combined for both Landsat 8 & 9, which have the same spectral bands. Cloudand shadow-free images of Landsat 8 and 9 data were extracted from GEE, then
compared with field observation data at different stands. Similar to section 5.3.1, daily
averaged foliage MC data at juvenile and mature stands were compared with remote

79
sensing indices (Figures 28 and 29), and duff and fine woody debris MC were
compared with remote sensing indices in open stands (Figures 30 and 31).

Figure 28: Foliage MC compared with five remote sensing indices at juvenile stands
with Landsat 8 & 9 data in 2022, where blue points are training data, green lines are the
linear regression based on blue points, and orange points are test data.

Figure 29: Foliage MC compared with five remote sensing indices at mature stands with
Landsat 8 & 9 data in 2022, where blue points are training data, green lines are the
linear regression based on blue points, and orange points are test data.
Model summaries and error statistics for remote sensing of foliage MC in juvenile
and mature forests are shown in Table 16. R² values range from 0.002 to 0.126 for
juvenile stands and from 0.032 to 0.297 for mature stands, while p values range from

80
0.0247 to 0.2890 for juvenile stands and from 0.0003 to 0.2740 for mature stands,
which indicates weak relationships. When analysing the results for each stand
individually, it can be observed that the GNDVI index had the highest R² value of 0.297
for the mature stands, and the GNDVI index had the highest R² value of 0.126 for the
juvenile stands. For the test data, the RMSE range for juvenile stands was between
18.01 and 26.89, with the lowest RMSE observed using GNDVI. In contrast, for mature
stands, the RMSE range was between 19.60 and 34.98, and the lowest RMSE was
observed using VARI.
Table 16: Statistical results of remote sensing indices with foliage MC at juvenile and
mature stands (2022 Landsat 8 & 9) where y is observed FMC.
Index Stand Type

R²

p

RMSE

Equation (train)

N train N test

NDVI

Juvenile

0.070 0.0999 24.01

y = 66.4*NDVI + 65.23

40

27

NDMI

Juvenile

0.002 0.0623 19.47

y = 5.32*NDMI + 110.34

40

27

NMDI

Juvenile

0.030 0.2890 26.89

y = -24.92*NMDI + 124.21

40

27

VARI

Juvenile

0.062 0.1200 19.61

y = -52.18*VARI + 95.09

40

27

GNDVI

Juvenile

0.126 0.0247 18.01

y = 186.36*GNDVI + 9.33

40

27

NDVI

Mature

0.032 0.2740 21.56

y = 78.39*NDVI + 57.80

39

26

NDMI

Mature

0.212 0.0032 24.31

y = -79.70*NDMI + 140.56

39

26

NMDI

Mature

0.090 0.0632 25.67

y = 128.12*NMDI + 25.79

39

26

VARI

Mature

0.216 0.0029 19.60

y = -174.08*VARI + 49.98

39

26

GNDVI

Mature

0.297 0.0003 34.98

y = 341.56*GNDVI – 61.71

39

26

Model summaries and error statistics for linear models were developed between
remote sensing indices and duff and fine woody debris MC at open stands (Figures 30
and 31) and are presented in Table 17. Similar to the findings in Section 5.3.1, the

81
relationship between the indices and duff and fine woody debris MC is relatively weak
compared to their relationship to foliage MC, indicating that these indices may not be
suitable for estimating woody debris MC. However, when using indices to estimate fine
woody debris MC, it has lower RMSE (17.57 – 44.40) compared to duff (59.10 – 69.32).

Figure 30: Duff MC compared with five remote sensing indices at open stands with
Landsat 8 & 9 data in 2022, where blue points are training data, green lines are the
linear regression based on blue points, and orange points are test data.

Figure 31: Fine woody debris MC compared with five remote sensing indices at open
stands with Landsat 8 & 9 data in 2022, where blue points are training data, green lines
are the linear regression based on blue points, and orange points are test data.

82

Table 17: Statistical results of remote sensing indices with duff and fine woody debris
MC at open stands, where FWD means fine woody debris, y is observed FMC (2022
Landsat 8 & 9).
Index

Fuel Type

R²

P

RMSE

Equation (train)

N train

N test

NDVI

duff

0.079

0.0374

69.32

y = -185.32*NDVI + 256.41

55

38

NDMI

duff

0.036

0.1650

59.93

y = -93.98*NDMI + 166.56

55

38

NMDI

duff

0.027

0.2290

64.59

y = -157.4*NMDI + 227.79

55

38

VARI

duff

0.006

0.5900

59.10

y = 37.99*VARI + 148.66

55

38

GNDVI

duff

0.013

0.4130

65.48

y = -120.98*GDNVI + 211.58

55

38

NDVI

FWD

0.004

0.6390

17.57

y = 17.87*NDVI + 16.66

55

38

NDMI

FWD

0.012

0.4250

44.40

y = -6.70*NDMI + 19.96

55

38

NMDI

FWD

0.008

0.5240

36.63

y = 21.19*NMDI + 10.56

55

38

VARI

FWD

0.004

0.6590

23.99

y = -11.25*VARI + 22.09

55

38

GNDVI

FWD

0.029

0.2640

18.90

y = 74.00*GNDVI – 11.41

45

31

When comparing the results between sections 5.3.1 (Sentinel 2, 2021) and 5.3.2
(Landsat 8 & 9, 2022), it can be seen that the R² values are generally higher, p-values
are generally lower in Table 16 and 17 than in Table 14 and 15. This suggests that the
relationship between the indices and foliage/duff MC is stronger and more statistically
significant when using Landsat 8 & 9 data in 2022 than Sentinel 2 in 2021.
As previously discussed in the Methods chapter, the occurrence of multiple
wildfires in central BC during 2021 resulted in significant smoke that may have impacted
the remote sensing data collected during that year. Additionally, the sample collection
dates in 2021 were only sometimes closely aligned with satellite collection dates, which
may have further contributed to potential errors in the data. To determine which method

83
is more reliable, the remote sensing index derived from Sentinel 2 data was compared
to the observed FMC data collected in 2022.

5.4.3 Estimating FMC with Sentinel 2 data in 2022.
As stated in section 4.4.1, after January 25, 2022, the Sentinel 2 harmonized
collection shifted data in new scenes to match the range of previous scenes. While
using the harmonized collection, the QA bands were ineffective in removing cloudy
pixels, so an NDVI filter was used instead. In 2022, sample collections around Smithers
and Prince George were timed to coincide with Sentinel 2 overpasses, resulting in
higher sample numbers than in 2021. Relationships between different remote sensing
indices and observed MC are shown in Figures 32 and 33, with the orange lines
indicating their best-fitting lines. Although there were more samples compared to data
from 2021 using Sentinel 2 and Landsat 8 and 9 data in 2022, it created more noise and
errors.

Figure 32: Remote sensing indices compared with foliage MC for juvenile stands with
Sentinel 2 data in 2022, where blue points are training data, green lines are the linear
regression based on blue points, and orange points are test data.

84

Figure 33: Remote sensing indices compared with foliage MC for mature stands with
Sentinel 2 data in 2022, where blue points are training data, green lines are the linear
regression based on blue points, and orange points are test data.
Statistical results from the 2022 Sentinel 2 MC models are given in Table 18. At
juvenile stands, GNDVI had the highest R² value (0.175) and the lowest p value (0.0004)
among all the indices, followed by VARI (R² = 0.139 and p = 0.0019) and NDMI (R² =
0.137 and p = 0.0021. For mature stands, NMDI had higher R² value (0.160) among all
the indices, followed by GNDVI (R² at 0.124), NDVI (R² at 0.110). Linear regression
models showed stronger relationships for juvenile stands than mature stands when
estimating foliage MC. The RMSE range for juvenile stands is between 16.97 – 22.87,
which is close to mature forests (15.50 to 24.94).
Table 18: Statistical results of remote sensing indices relationships with foliage MC at
juvenile and mature stands (Sentinel 2, 2022), where y is observed foliage MC.
Index Stand Type

R²

P

RMSE

Equation (train)

N train N test

NDVI

Juvenile

0.083

0.0181

17.56

y = 42.31*NDVI + 82.85

67

45

NDMI

Juvenile

0.137

0.0021

16.97

y = 105.56*NDMI + 66.97

67

45

NMDI

Juvenile

0.003

0.1858

21.22

y = -27.35*NMDI + 126.85

67

45

85

VARI

Juvenile

0.139

0.0019

21.82

y = -8.31*VARI + 99.8

67

45

GNDVI

Juvenile

0.175

0.0004

22.87

y = 58.23*GNDVI + 79.51

72

45

NDVI

Mature

0.110

0.0058

17.88

y = 53.63*NDVI + 80.34

68

46

NDMI

Mature

0.160

0.0006

20.81

y = 149.52*NDMI + 63.17

69

46

NMDI

Mature

0.006

0.4144

23.41

y = -26.37*NMDI + 126.7

69

46

VARI

Mature

0.092

0.0119

15.50

y = -18.03*VARI + 97.73

68

46

GNDVI

Mature

0.124

0.0033

24.94

y = 40.42*GNDVI + 82.97

68

46

Relationships between duff and fine woody debris moisture contents in open
stands were examined with Sentinel 2 indices in 2022 (Figures 34 and 35). Similar to
the results from Sentinel 2 data in 2021 and Landsat 8/9 data in 2022, the relationship
between the indices and fine woody debris MC is generally weak (Table 19). However,
there is some promise in estimating duff MC with NDVI and GNDVI indices, which have
relatively higher R² values (0.210 and 0.157) and lower p-values (<0.05) (Table 19).
Although there are no clear relationships between woody debris MC and remote
sensing indices, RMSE values for fine woody debris models (21.36 to 56.56) are lower
than for duff (69.26 to 85.88) (Table 19).

86

Figure 34: Remote sensing indices compared with duff MC at open stands with Sentinel
2 data in 2022, where blue points are training data, green lines are the linear regression
based on blue points, and orange points are test data.

Figure 35: Remote sensing indices compared with woody debris MC at open stands
with Sentinel 2 data in 2022, where blue points are training data, green lines are the
linear regression based on blue points, and orange points are test data.

87
Table 19: Statistical results of remote sensing index relationships with fine woody debris
and duff MC at open stands (2022 Sentinel 2), where FWD means fine woody debris,
and y is observed FMC.
Index

Fuel Types

R²

P

RMSE

Equation (train)

N Train N Test

NDVI

duff

0.210 <0.0001 85.22

y = -243.00*NDVI + 272.87

78

52

NDMI

duff

0.048

0.0549

73.49

y = -260.71*NDMI +187.38

78

52

NMDI

duff

0.072

0.0174

85.88

y = 383.11*NMDI – 69.23

78

52

VARI

duff

0.045

0.0629

72.80

y = 77.74*VARI + 180.20

78

52

GNDVI

duff

0.157

0.0003

69.26

y = -313.80*GNDVI + 301.72

78

52

NDVI

FWD

0.094

0.0064

56.56

y = -47.42*NDVI +50.36

78

52

NDMI

FWD

0.031

0.1250

45.36

y = -73.29*NDMI + 39.98

78

52

NMDI

FWD

0.060

0.0303

39.92

y = 194.32*NMDI – 76.61

78

52

VARI

FWD

0.002

0.1336

21.36

y = 6.23*VARI + 36.50

78

52

GNDVI

FWD

0.015

0.2810

36.41

y = -42.25*GNDVI + 53.55

78

52

Taken as a whole, these results indicate that in 2021 and 2022, for juvenile
stands, the GNDVI, NDVI, and NMDI had a stronger correlation with field observations
and Sentinel 2 data, as evidenced by its higher R² values and lower p-values. NMDI had
stronger relationships and correlations in 2021, although its low R² and high p-value in
2022 may make it unreliable. No solid relationships were found between observed MC
and remote sensing indices in the 2021 data for mature stands. However, in 2022,
NDMI, GNDVI, and NDVI both had relatively high R² values and lower p-values.
Furthermore, at open stands, the correlation between remote sensing indices with duff
and fine woody debris MC is relatively weak in 2021 and 2022 with Sentinel 2 data.

88
However, NDVI and GNDVI showed stronger relationships and correlations in 2022
when estimating duff at open stands.
For Landsat 8 & 9 data in 2022 for juvenile stands, only GNDVI had a higher R²
value and a lower p-value, while at mature stands, GNDVI, NDVI and GNDVI show
statistically significant relationships. All indices had poor statistical results when using
Landsat 8 & 9 data to estimate duff and fine woody debris MC.
In conclusion, GNDVI appears to work best to estimate foliage MC in juvenile
and mature stands, followed by NDVI and NMDI. In open stands, GNDVI and NDVI
have better statistical results than other indices when estimating duff MC, and no index
worked well with estimating woody debris MC. In the following sections, the use of
GNDVI as an index of pre-fire FMC and comparing it with estimates of the burn severity
will be discussed and explored.

5.5 Burn severity, moisture, indices, and FMC
Distributions of dNBR and RBR burn severities for open, juvenile, and mature
stands in the Plateau Complex Fire are shown in Figures 36 and 37. Results indicate
that juvenile stands had lowest fire intensities/severities, followed by mature forests, and
open stands (high intensity/severity) (Figures 36 and 37)
The juvenile blocks (blue polygons) had lower burn severity than mature and
open stands, as measured by dNBR and RBR (Figures 36 and 37). It is reasonable to
find some low burn severity patches not related to underlying indices/fuel differences, as
they may have burned under cooler conditions (e.g., at night), or the fire has stopped or
changed its direction while spreading. For example, in Figures 36 and 37, most open
forests (recent clearcuts) have experienced high severity fire, but the open patches
surrounded by juvenile forests only had low burn severity. Before comparing the indices
and burn severities, it is essential to understand the fire behaviour and test the accuracy
of using remote sensing to estimate burn severity.

89

Figure 36: dNBR for different stands on 6th July 2017 at Plateau Complex, centred on
52.94105 N, -124.01571 W.

90

Figure 37: RBR value for different stands on 6th July 2017 at Plateau Complex, centred
on 52.94105 N, -124.01571 W.
Regardless of using dNBR and RBR, juvenile stands are mainly unburned
(majority of dNBR <100 and RBR <35, Figures 38- 39). At the same time, mature
forests showed a wide distribution of unburned to higher burn severity (higher than 660
for dNBR and 298 for RBR, Figures 38- 39) values. Open stands had higher dNBR and
RBR in general, which indicates the majority of the recently logged open forest has
burned (Figures 38- 39).

91

Figure 38: Histogram of dNBR for different stand types, where dashed lines represent
median dNBR.

Figure 39: RBR histogram in different stands, where dashed lines represent median
RBR.
Pre-fire NDMI, NMDI, NDVI, VARI and GNDVI at different stands were used to
compare with dNBR and RBR, as their relationships to the FMC had been determined in
the previous section. The relationships between remote sensing indices and dNBR/RBR

92
in different stands are shown in Figures 40 to 45. As there are more than 49,000 data
points, hexbin plots were used to visually investigate their relationships, where darker
colour means more dense points/points overlay. There is a wide range of weak
relationship when comparing burn severities and indices, as the study area is a mix of
burned and unburned forests.
Most open forests within the scene selected in the Plateau Complex Wildfire
have been burned, as shown by relatively high dNBR/RBR. Mature forests have
moderate burn severity values compared to the other two stand types. For juvenile
forests, within the scene selected, the majority of juvenile forests was unburned. The
relationship between indices and burn severities is unclear in juvenile and open forests.
However, despite the noise, all indices except VARI in mature stands have an inverse
relationship to burn severities (Figures 40 and 43): as the indices decreased, the burn
severity increased.
A

D

B

C

E

Figure 40: A) NDVI, B) NDMI, C) NMDI, D) VARI and E) GNDVI vs dNBR for mature
forests, with their frequency distributions

93

A

B

D

E

C

Figure 41: A) NDVI, B) NDMI, C) NMDI, D) VARI and E) GNDVI vs dNBR for open
forests, with their frequency distributions

94

A

B

D

E

C

Figure 42: A) NDVI, B) NDMI, C) NMDI, D) VARI and E) GNDVI vs dNBR for juvenile
forests, with their frequency distributions.

95

A

B

D

E

C

Figure 43: A) NDVI, B) NDMI, C) NMDI, D) VARI and E) GNDVI vs RBR for mature
forests, with their frequency distributions.

96

A

D

C

B

E

Figure 44: A) NDVI, B) NDMI, C) NMDI, D) VARI and E) GNDVI vs RBR for open
forests, with their frequency distributions.

97

A

B

D

C

E

Figure 45: A) NDVI, B) NDMI, C) NMDI, D) VARI and E) GNDVI vs RBR for juvenile
forests, with their frequency distributions.
The relationship between remote sensing indices, such as NDMI and FMC, was
investigated for different stand types in previous sections. Results indicate a relationship
between the two in various stands; however, the remote sensing indices in different
stands do not relate to subsequent burning severity in general.
Both NDMI and GNDVI have relatively higher R² and lower p values when used
to estimate foliage MC at juvenile and mature forests (Table 19). These relationships
were used in juvenile and mature stands to estimate their FMC in different stands.
Figures 46 and 47 shows estimated FMC at mature and juvenile forest computed by
equations from Table 19. Juvenile forest has higher FMC in general than mature forest,
regardless of using NDMI or GNDVI equations. For FMC calculated by NDMI (Figure
46), most of the juvenile forest have FMC 100-120 % as well, however, at mature forest,
FMC is between 60-100%, which is generally lower than juvenile. In Figure 47
(computed by GNDVI), most of the juvenile forest have FMC between 100-120%, while
most of mature forest have FMC between 80-100 %.

98

Figure 46: FMC calculated by NDMI for juvenile and mature stand on 6th July 2017 at
Plateau Complex, centred on 52.94105 N, -124.01571 W.

99

Figure 47: FMC calculated by GNDVI at juvenile and mature stand on 6th July 2017 at
Plateau Complex, centred on 52.94105 N, -124.01571 W.
In Figure 48 and 49, FMC estimated from NDMI and GNDVI at juvenile and
mature strands were compared with the burn severities. Similar to NDMI/ GNDVI and
burn severity in Figure 40 and 43, despite the noise, FMC calculated by NDMI and
GNDVI at mature stands have an inverse relationship to burn severities; as the FMC
decreased, the burn severity increased. There is no clear relation between FMC at
juvenile stands and burn severities, as most of juvenile stands were unburned.
Different stands with similar remote sensing indices and FMC values tend to
have a wide range of burn severity (Figures 40 to 45, Figures 48 and 49), which
suggests that remote sensing indices and FMC alone is not determining whether a
stand has been burned.

100

A

C

B

D

Figure 48: A). FMC calculated by NDMI at mature stands, B). FMC calculated by
GNDVI at mature stands, C). FMC calculated by NDMI at juvenile stands, D). FMC
calculated by GNDVI at juvenile stands compared with dNBR.

101

A

C

B

D

Figure 49: A). FMC calculated by NDMI at mature stands, B). FMC calculated by
GNDVI at mature stands, C). FMC calculated by NDMI at juvenile stands, D). FMC
calculated by GNDVI at juvenile stands compared with RBR.

102

6. Discussion
6.1 Field observations
As mentioned in section 5.1, the moisture content (MC) of duff, fine woody debris,
and foliage were compared in different stand types. In 2021 and 2022, duff had a
significantly higher daily averaged MC in mature stands, followed by juvenile and open
stands. For fine woody debris, the daily averaged MC was similar in mature and juvenile
stands, with no significant difference (p=0.851). However, open stands had a
significantly lower daily averaged MC than juvenile stands. When looking at foliage MC,
open forests had a significantly higher daily averaged MC, while mature and juvenile
forests had similar daily averaged MC without significant differences.
Mature forests shed more needles/foliage, creating a thicker layer of litter that
can trap moisture and slow down evaporation, resulting in higher MC in the duff layer
(Harmon et al., 1986). Fine woody debris in mature conifer forests may have undergone
more decomposition than in juvenile conifer forests, resulting in a higher overall
moisture content, however, this can be offset by the accumulation of new fine woody
debris in mature forests (Brown, 2003; Harmon et al., 1996). The major reason why
open stands have lower duff and fine woody debris MC could be the lack of canopy
cover. Mature and juvenile forests have denser canopy cover that shades the forest
floor and reduces the rate of evaporation, helping to maintain moisture levels in the fine
woody debris and duff layer (Wotton et al., 2005).
At open stands, foliage was collected from actively growing trees younger than
10 years old. Active growth in younger trees might result in higher moisture content as
they require a high level of water intake (Lambers et al., 1998).
Although field observations of FMC yield direct estimates of moisture levels in
forest fuels, it is time-consuming and costly (Caccamo et al., 2011). Collecting samples
on the day of satellite image acquisition is desirable to correlate field measurements
directly with satellite data; however, collecting samples from all six locations used in this
study in one day was not possible. Specific errors and outliers occurred throughout the

103
summer collection, such as negative values of FMC caused by recording errors. Several
reasons might have contributed to the defects and inconsistencies, including.
1. Transportation: In the beginning, the team in Prince George used sealed
sandwich bags to retain the samples, with water frequently left on the bottom
before measurement (lower moisture content than expected).
2. Need for more experience and canopy access for sample collection: Failure to
correctly label tree foliage species may have led to data analysis errors. Samples
collected primarily from lower tree branches in juvenile and mature stands are
also controversial to compare with remote sensing data as their exposed
canopies were too high to reach.
3. Calculation/ measurement errors: The loss of samples happened several times
while measuring and transferring in and out of the oven. Collecting samples while
raining, or immediately following a rain event, may also have caused
measurement errors if not properly labelled to allow exclusion.
An improved sample collection procedure could provide a more reliable
observation database to compare with both empirical models and remote sensing
indices. Future sample collection should use different methods such as drones or
shotguns to collect higher foliage exposed to sunlight (Charron et al. 2020).

6.2 Empirical Models to Estimate Fuel Moisture Content
Weather conditions can vary across geographic regions and forest types, and the
current fire weather stations can only accurately represent a limited range of forest
areas. Results presented in section 5.2 suggest that the installation of in-stand sensors
may yield a more accurate estimate of FMC, but a full revision of the FWI model, or the
development of new empirical models to estimate FMC that considers in-stand data
may be required. The operation of in-stand sensors can be costly and labour-intensive:
sensors installed in remote forest areas require maintenance and regular visits to
prevent data loss from wildlife and extreme weather. Alternatives to in-stand
meteorological observations include interpolation of weather conditions from multiple
nearby fire weather stations, weather satellites and high-resolution dynamically
downscaled meteorological fields (Mandel et al., 2011). Weather satellites have the

104
potential to be used to monitor fuel moisture content remotely. However, their finest
resolution is from the United States Department of Defense's Meteorological Satellite
(DMSP) (2.7 km), Advanced Baseline Imager (ABI) (500m), JAXA Himawari Monitor
(500 m), and which is too low to monitor different forest stands (Mohsin Butt, 2013;
National Oceanic and Atmospheric Administration, 2021; Japan Aerospace Exploration
Agency, 2015).
Over the course of the fieldwork conducted for this research, some sensors were
damaged by animals, causing data loss and equipment damage. Methods such as
spraying anti-chew spray and surrounding sensors with metal/ plastic chicken wire have
been used to protect the equipment.
Due to the lack of equipment and the challenge of setting them up in dense
forests, most of the weather sensors used in this research are quite different from the
standard fire weather station. At all open sites around Smithers and Prince George,
although temperature and relative humidity were measured at a screen level (1.5
metres), wind speed was collected at a height of approximately 2 metres instead of 10
metres. In juvenile and mature stands, the wind speed sensor was installed at a height
of approximately 2 metres, and temperature and relative humidity were collected around
0.30 m above the ground. These measurement heights may be more relevant to surface
fuel conditions but make comparisons with standard weather and FWI data difficult.
The comparison of different FMC models in section 5.2 revealed that while
updating local weather and fuel conditions improved the models' performance, they
could have worked more consistently. The updated models performed better for open
sites than for mature and juvenile forest stands. The weather data for calculating and
converting drying and wetting rates in the DMC and FFMC models were based on
empirical relationships derived for pine forests in boreal and eastern Canada.
Differences in forest types and weather conditions can result in differing empirical drying
and wetting rates, leading to potential errors when using these rates in central BC.
Given that the FWI model was developed and completed in the 1960 to 1970s (Van
Wagner, 1987), it is necessary to update the empirical model to include fuel collection
data from different forests and regions. As research and operational experience
progress, fire management challenges change, and technology advances, along with

105
the expansion of data on the fire environment, the Canadian Forest Service Fire Danger
Rating System (CFFDRS) must continue to update and evolve, despite previous
updates (Canadian Forest Service Fire Danger Group, 2021).
Empirical models had greater skill and reduced errors in the estimation of fine
woody debris MC than for duff MC. This is likely due to the high variability of observed
duff MC. One of the critical reasons for the high variability in the MC of duff is its high
porosity, which allows for rapid absorption and release of moisture in response to
environmental conditions (Stocks et al., 2004). Factors such as temperature, humidity,
wind, and solar radiation influence the moisture absorption and loss rate and can lead to
significant variability in duff MC over short periods and short distances (Deeming et al.,
1997). Additional reasons behind this variability could be the difficulty of collecting duff
in open and juvenile stands or improper identification of the duff layer, as mentioned in
section 6.1.

6.3 Remote Sensing of FMC
In section 5.3, none of the remote sensing indices tested showed a strong
relationship with the FMC, which can be attributed to several reasons. Firstly, the time
series of remote sensing data were interpolated using lowess interpolation methods, but
the accuracy of this approach depends on the frequency and quality of data. During the
wildfires in central BC in the summer of 2021, clouds and smoke significantly decreased
the number of usable image capture days. Figure 50 shows the number of useable
images from Sentinel 2 at the different sample sites in 2021. The significant gaps in
image capture likely led to inaccuracies in the remote sensing index values extracted
from the lowess function. Fortunately, in 2022, there were fewer wildfires, providing
more frequent data than in 2021. However, the harmonised 2022 data resulted in the
QA bands not working when filtering out clouds and cirrus clouds. This study used a
NDVI threshold to remove cloud and shadow pixels, however, using NDVI to remove
cloud/shadow pixels may introduce errors. Although matching sample collection days
and satellite visit days can minimise the errors caused by the lowess function, due to
labour limitations, MC samples could not be collected for all clear satellite visit days,
which made it still necessary to use the lowess function for the 2022 data. Running a

106
smoothing function through the remote sensing index time series could also introduce
errors, as it may not reflect the actual indices at the sample collection dates accurately.
Further research could test the parameters of the lowess smoothing functions, but
without more field observations and greater image frequency it will be a challenge to
accurately compare field and remote sensing data.

Figure 50: Usable NDMI indices in 6 sample collection sites in 2021 summer (late May
to September)
The relationship between FMC and remote sensing indices should also consider
factors such as the type of foliage (pine, spruce, or fir), sun or shade exposure, and the
foliage's height. However, in this study, only three samples were taken per stand per
visit, and the challenge of keeping the foliage type and exposure consistent was not
considered. Although the daily averaged foliage MC for different foliage types also don’t
have significant different as well as the same stand, it might still be better to keep using
the same foliage type at same stands. As a result, the comparison between remote
sensing indices and FMC is not fully representative. In future studies, using exposed
fuel data/ foliage from the canopy top could provide more representative results.
Satellites such as Sentinel 2 and Landsat 8 & 9, with a resolution of 30 metres, can
introduce inaccuracies in comparing remote sensing indices and FMC as each pixel

107
includes non-foliage components such as trails, trees, and ground (as shown in Figure
3). Additionally, the frequency and timing of satellite visits can also impact the accuracy
of results. For example, Landsat 8/9 visited the sample collection sites around Prince
George and Smithers at around 10 a.m., but many samples were taken at 2 - 3 pm due
to transportation and labour constraints. To obtain more accurate results, future studies
could use drone-based remote sensing data, which has a higher resolution, is less
impacted by clouds, and is closer to the sample collection time.
Additionally, the regression models used to estimate FMC from remote sensing
indices are sensitive to the subset of points used for model training. Further research
into this sensitivity is required.

6.4 Burn Severity Case Study
A portion of the 2017 Plateau Complex Wildfire was explored as a case study to
examine potential relationships between wildfire severity and remote sensing indices
related to pre-fire fuel moisture content. While harvest data provides mapped areas of
open stands, and VRI data defined most of the forest with their forest age, there were
still some forests not defined. Clouds, shadows, and smoke can also decrease the
chance of getting pre-fire imagery prior to the wildfire event, which may lead to false
pre-fire forest conditions.
The results shown in section 5.4 indicated that there is no strong relation
between remote sensing indices and burn severity ratios. In general, most open,
recently logged stands had high burn severities, many juvenile forest stands remained
unburned, while mature forests have an intermediate distribution of burn severity.
At the end of summer in 2021, ground-based severity assessments were
conducted Plateau Complex wildfire sites (data not presented). To evaluate burn
severity results between remote sensing and on-site observation, 15 of the unburned
forests (fire island) and 17 burned forests were visited. When visiting unburned
areas/fire islands, it was common to notice that their edges had a relatively higher burn
severity than the centre of the juvenile forest stand. From burn severity assessments
and fieldwork observation, wildfires are not just suddenly stopped when they meet the

108
juvenile stands, they burn the edges of the stands, and their spreading slows as they go
deeper into the forest. Evidence of low intensity fire close to the ground, such as
scorched tree bases, less moss and recently established fireweed (Epilobium
angustifolium), was found frequently within the fire refugia, even far from the edge of the
forest. Open stands and very young juvenile stands, for example, can have a higher
burn severity potential due to their high woody debris components after recent logging
(Brown 2003, Lindenmayer et al. 2009).
Section 5.1 showed that FMC was similar in juvenile and mature forests in Prince
George and Smithers in 2021 and 2022, and that weak relationships exist between
remote sensing indices and FMC. GNDVI, NDMI, and NDVI indices showed potential in
estimating future wildfire severity in mature stands (Figures 44 and 47). However, no
strong relations were found between pre-fire FMC and burn severities, which suggests
that FMC is not a decisive factor for forest survival during wildfires. Linear and nonlinear models were not examined in this research due to the spread of the data. Future
work should attempt to identify the form of the relation between burn severity and
moisture/greenness indices.
The high forest density in juvenile forests may also play a role in reducing the
spread of fire by decreasing wind speed. In section 4.3.3, wind speed was found to be
significantly lower than mature and open stands, which may prevent fire spread and
create fire refugia. This highlights the complex interplay between various factors that
can influence the impact of wildfires on forests, and the importance of considering a
range of factors when evaluating the resilience of forests to fire.

7. Conclusions
This research compares empirical models and remote sensing indices for
predicting forest duff, fine woody debris, and foliage MC values measured in central BC.
Burn severities and their relation to pre-fire remote sensing indices were also analysed
in different stands from the wildfire case study to understand the fire behaviour in fire
refugia.

109
The FWI and its locally modified versions were tested in this research. By
comparing the different approaches in DMC and FFMC models, it was found that using
local fuel limits and weather station data resulted in better estimates of duff and woody
debris MC. However, the cost of weather stations and sensors and various wetting and
drying rates in the model made it difficult for any of the models to work consistently well.
When comparing the performance of FWI models in different stand types, open
(recently logged) stands had lower errors than juvenile and mature conifer stands, as
the model was designed to estimate MC with weather data collected in open stands in
Eastern Canada. Although the model was designed to and intended to be
representative of natural forests, the mature and juvenile forests in central BC are
different than forests in Eestern Canada and US. In general, FFMC and its locally
updated version worked better than DMC at predictive local MC of fine woody fuels and
duff, possibly due to the greater variability of observed duff MC (14.38 – 439.27 %
compared to fine woody debris 2.17 – 228.38 %).
After comparing observed MC and remote sensing indices at the different stands
and sites, a few relationships can be highlighted. Firstly, Sentinel-2 data yielded
stronger relationships between moisture content and remote sensing indices. GNDVI,
NDMI, and NDVI had higher R² values (higher than 0.1) and lower p-values (less than
or equal to 0.005) when estimating foliage MC in juvenile and mature stands, compared
to NDVI and GNDVI. When estimating duff MC in open stands, NDVI and GNDVI had
better statistical results (higher R² and lower p-value) than other remote sensing indices.
In contrast, almost none of the remote sensing indices had a clear relationship with fine
woody debris MC in open stands. In general, the results of using remote sensing indices
to estimate FMC were not convincing and accurate, with the highest R² being 0.175
(GNDVI) for foliage in mature forests, 0.160 (NDMI) for juvenile forests, 0.210 (NDVI)
for duff, 0.094 (NDVI) for fine woody debris in open stands. Due to the impact of satellite
visit frequency, band resolution, weather, and the quality and quantity of observation
data, it is unlikely that strong relationships between remote sensing indices and FMC
can be obtained.
A case study of the 2017 Plateau Complex fire in central British Columbia was
used to examine pre-and post-fire forest conditions through remote sensing. The results

110
showed that juvenile stands in the Plateau complex fire had lowest burn severity in
general, followed by mature forests and open stands. Remote sensing indices related to
moisture or greenness could not be used to determine fire severity in juvenile stands but
did provide information on the fire severity in mature stands. By using NDMI and GNDVI
equations, juvenile forests tended to have a higher moisture content level than mature
forests, however, the relations of FMC between FMC and dNBR/RBR is still
undetermined. Thus, FMC might not be a conclusive factor in the observation that
juvenile forests can function as fire refugia. The relationship between remote sensing
indices and burn severity is still being determined when looking at a broader study area.
Juvenile stands had significantly lower in-stand wind speeds than both mature and open
stands (Figure 18), and this should be investigated further for understanding fire refugia
and wildfire behaviour in central British Columbia.

8. References
Anderson, H. E. 1982. Aids to determining fuel models for estimating fire behavior. USDA Forest
Service, Intermountain Forest and Range Experiment Station.
Bao, W., Yu, F., Chang, Y., Guo, M., Zhou, F., & Zhong, C. 2022. Estimation of fuel load using
remote sensing data in Hulunbuir Grassland. Natural Hazards Research (pp. 375-383).
Blackmarr, W. H., & Flanner, W. B. 1968. Seasonal and diurnal variation in moisture content of six
species of pocosin shrubs (GTR- 033). US Department of Agriculture, Forest Service, Southeastern
Forest Experiment Station.
Brown, J. K. 2003. Coarse woody debris: managing benefits and fire hazard in the recovering forest.
Gen.Tech. Rep. RMS-GTR-105, USDA Forest Service, Ogden, UT. 16 p.
Burton, P. 2022. understanding the effects of forest management and prior disturbances on forest fire
risk and behavior, Natural Resources Canada year report, Victoria, B.C, 12p.
Burton, P., C. Mahood, A. Boxwell, S. Taylor, L. McCulloch, and M. Dale. 2019. Why have recent
mega-fires in British Columbia left some established plantations unburned? North American Forest
Ecology Workshop, 24 June 2019, Flagstaff, Arizona.
Butt, M. J. 2012. Estimation of light pollution using satellite remote sensing and geographic
information system techniques. GIScience & Remote Sensing, 49(4), 609-621.
Byram, G. M. 1959. Combustion of forest fuels. In Forest fire: control and use (pp. 61-89). McGrawHill
Caccamo, G., Chisholm, L. A., Bradstock, R. A., & Puotinen, M. L. 2011. Assessing the sensitivity of
MODIS to monitor drought in high biomass ecosystems. Remote Sensing of Environment, 115(10),
2626-2639.

111
Caccamo, G., Chisholm, L. A., Bradstock, R. A., Puotinen, M. L., & Pippen, B. G. 2011. Monitoring
live fuel moisture content of heathland, shrubland and sclerophyll forest in south-eastern Australia
using MODIS data. International Journal of Wildland Fire, 21(3), 257-269.
Camacho, J. 2014. Visualizing big data with compressed score plots: approach and research
challenges. Chemometrics and Intelligent Laboratory Systems, 135, 110-125.
Canadian Forest Service Fire Danger Group. 2021. The Vision for the Next Generation of the
Canadian Forest Fire Danger Rating System. Information Report GLC-X-26. Canadian Forest Service,
Sault Ste. Marie, ON. 56 p
Ceccato, P., Flasse, S., & Gregoire, J. M. 2002. Designing a spectral index to estimate vegetation
water content from remote sensing data: Part 2. Validation and applications. Remote sensing of
Environment, 82(2-3), 198-207.
Charron, G., Robichaud-Courteau, T., La Vigne, H., Weintraub, S., Hill, A., Justice, D., ... & Lussier
Desbiens, A. 2020. The DeLeaves: a UAV device for efficient tree canopy sampling. Journal of
Unmanned Vehicle Systems, 8(3), 245-264.
Chuvieco, E. 2008. Satellite observation of biomass burning. In Earth Observation of Global Change
(pp. 109-142). Springer, Dordrecht.
Chuvieco, E., Aguado, I., Cocero, D., & Riano, D. 2003. Design of an empirical index to estimate fuel
moisture content from NOAA-AVHRR images in forest fire danger studies. International Journal of
Remote Sensing, 24(8), 1621-1637.
Chuvieco, E., Aguado, I., Yebra, M., Nieto, H., Salas, J., Martín, M. P., ... & Zamora, R. 2010.
Development of a framework for fire risk assessment using remote sensing and geographic
information system technologies. Ecological modeling, 221(1), 46-58.
Chuvieco, E., González, I., Verdú, F., Aguado, I., & Yebra, M. 2009. Prediction of fire occurrence from
live fuel moisture content measurements in a Mediterranean ecosystem. International Journal of
Wildland Fire, 18(4), 430-441.
Cleveland, W. S. 1979. Robust locally weighted regression and smoothing scatterplots. Journal of the
American Statistical Association, 74(368), 829-836.
Cohen, W. B., Spies, T. A., & Fiorella, M. 1995. Estimating the age and structure of forests in a multiownership landscape of western Oregon, USA. International Journal of Remote Sensing, 16(4), 721746.
Danson, F. M., & Bowyer, P. 2004. Estimating live fuel moisture content from remotely sensed
reflectance. Remote Sensing of Environment, 92(3), 309-321.
Data Catalogue. 2022. Retrieved January 28, 2023, from
https://catalogue.data.gov.bc.ca/dataset/harvested-areas-of-bc-consolidated-cutblocksDe Groot, W. J. 1998. Interpreting the Canadian forest fire weather index (FWI) system. In Proc. of
the Fourth Central Region Fire Weather Committee Scientific and Technical Seminar.
Deeming, J. E., Burgan, R. E., & Cohen, J. D. 1997. The National Fire Danger Rating System - 1978.
General Technical Report INT-GTR-341, USDA Forest Service, Intermountain Forest and Range
Experiment Station.

112
Dennison, P. E., Moritz, M. A., & Taylor, R. S. 2008. Evaluating predictive models of critical live fuel
moisture in the Santa Monica Mountains, California. International Journal of Wildland Fire, 17(1), 1827.
Dunn, C. J., & Bailey, J. D. 2016. Tree mortality and structural change following mixed-severity fire in
Pseudotsuga forests of Oregon’s western Cascades, USA. Forest Ecology and Management, 365,
107-118.
Flannigan, M. D., Stocks, B. J., & Wotton, B. M. 2000. Climate change and forest fires. Science of the
total environment, 262(3), 221-229.
Fonseca, L., & Andrade, P. 2019. Cloud removal from Sentinel-2 imagery for land use classification
using the NDVI threshold method. European Journal of Remote Sensing, 52(1), 357-368.
Fosberg, M. A. 1971. Fine herbaceous fuels in fire-danger rating (Vol. 185). Rocky Mountain Forest
and Range Experiment Station, Forest Service, US Department of Agriculture.
Furlaud, J. M., Prior, L. D., Williamson, G. J., & Bowman, D. M. 2021. Fire risk and severity decline
with stand development in Tasmanian giant Eucalyptus forest. Forest Ecology and Management, 502,
119724.
Gao, B. C. 1996. NDWI—A normalized difference water index for remote sensing of vegetation liquid
water from space. Remote sensing of environment, 58(3), 257-266.
Gao, J., Chen, S., & Wang, J. 2018. Assessment of soil moisture in a semi-arid region using remote
sensing data. Remote Sensing, 10(1), 21.
Gitelson, A. A., Kaufman, Y. J., Stark, R., & Rundquist, D. 2002. Novel algorithms for remote
estimation of vegetation fraction. Remote sensing of Environment, 80(1), 76-87.
Google Earth Engine. (2022). Copernicus Sentinel-2 Surface Reflectance Harmonized [Data set].
Retrieved from https://developers.google.com/earthengine/datasets/catalog/COPERNICUS_S2_SR_HARMONIZED
Government of Canada, Canadian Forest Service. (n.d.). Fire Weather Index (FWI). Retrieved from
https://cwfis.cfs.nrcan.gc.ca/background/summary/fwi
Government of British Columbia. (n.d.). Harvested Areas of BC: Consolidated Cutblocks. Retrieved
from https://catalogue.data.gov.bc.ca/dataset/harvested-areas-of-bc-consolidated-cutblocks
Government of British Columbia. (Year). Vegetation Resources Inventory (VRI) Spatial
Datasets .Open Government of Canada. Retrieved from
https://open.canada.ca/data/en/dataset/02dba161-fdb7-48ae-a4bb-bd6ef017c36d?wbdisable=true
Hartford, R. A., & Frandsen, W. H. 1992. When it's hot, it's hot... or maybe it's not!(Surface flaming
may not portend extensive soil heating). International Journal of Wildland Fire, 2(3), 139-144.
Japan Aerospace Exploration Agency. 2015 PTree User's Guide. Earth Observation Research Center.
Retrieved from https://www.eorc.jaxa.jp/ptree/userguide.html.
Jin, S., Tian, L., & Chen, Y. 2021. Monitoring vegetation growth and its response to climate change in
the Yellow River Basin, China. Environmental Science and Pollution Research, 28(1), 19-30.

113
Jolly, W. M., Cochrane, M. A., Freeborn, P. H., Holden, Z. A., Brown, T. J., Williamson, G. J., &
Bowman, D. M. 2015. Climate-induced variations in global wildfire danger from 1979 to 2013. Nature
Communications, 6(1), 1-11
Kane, V. R., Cansler, C. A., Povak, N. A., Kane, J. T., McGaughey, R. J., Lutz, J. A., ... & North, M. P.
2015. Mixed severity fire effects within the Rim fire: relative importance of local climate, fire weather,
topography, and forest structure. Forest Ecology and Management, 358, 62-79.
Karki, S., & Chaudhary, R. P. 2018. The Impact of Forests on Local Climate: A Study of Temperature
and Precipitation in Nepal. Forests, 9(8), 481
Key, C., & Center, G. F. S. 2006. Evaluate sensitivities of burn-severity mapping algorithms for
different ecosystems and fire histories in the United States. Final Report to the Joint Fire Science
Program.
Key, C. H., & Benson, N. C. 2006. Landscape assessment (LA). FIREMON: Fire effects monitoring
and inventory system, 164, LA-1.
Klinka, K., Krajina, V. J., CESIS Forestry Sciences, UBC, Vancouver, B. C., & Carter, R. E. 1999.
Indicator plants of coastal British Columbia. University of British Columbia Press.
Krawchuk, M. A., Haire, S. L., Coop, J., Parisien, M. A., Whitman, E., Chong, G., & Miller, C. 2016.
Topographic and fire weather controls of fire refugia in forested ecosystems of northwestern North
America. Ecosphere, 7(12), e01632.
Krawchuk, M. A., Moritz, M. A., Parisien, M. A., Van Dorn, J., & Hayhoe, K.. 2009. Global
pyrogeography: the current and future distribution of wildfire. PloS one, 4(4), e5102.
Landsat NASA. 2021. Landsat 9 Bands | Landsat Science. Landsat Science | a Joint NASA/USGS
Earth Observation Program. Retrieved from https://landsat.gsfc.nasa.gov/satellites/landsat-9/landsat9-bands/
Lawson, B. D., & Armitage, O. B. 2008. Weather guide for the Canadian forest fire danger rating
system.
Lindenmayer, D. B., Hunter, M. L., Burton, P. J., & Gibbons, P. 2009. Effects of logging on fire
regimes in moist forests. Conservation Letters, 2(6), 271-277.
Mandel, J., Beezley, J. D., Coen, J. L., & Kim, M. 2011. WRF-Fire: Coupled weather–wildland fire
modeling with the Weather Research and Forecasting Model. Journal of Applied Meteorology and
Climatology, 50(2), 344-367.
Masek, J. G., Wulder, M. A., Markham, B., McCorkel, J., Crawford, C. J., Storey, J., & Jenstrom, D. T.
2020. Landsat 9: Empowering open science and applications through continuity. Remote Sensing of
Environment, 248, 111968.
McGee, Tara, Bonita McFarlane, and Cordy Tymstra 2015. "Wildfire: a Canadian perspective."
Wildfire hazards, risks and disasters. Elsevier,. 35-58.
Meigs, G. W., Dunn, C. J., Parks, S. A., & Krawchuk, M. A. 2020. Influence of topography and fuels
on fire refugia probability under varying fire weather conditions in forests of the Pacific Northwest,
USA. Canadian Journal of Forest Research, 50(7), 636-647.
Miller, B. W., Lutz, J. A., Holden, Z. A., & Morgan, P. 2019. Mapping burn severity using Landsat and
Sentinel-2 data: A comparison of dNBR and RdNBR. Remote Sensing, 11(11), 1307.

114
Miller, J. D., Knapp, E. E., Key, C. H., Skinner, C. N., Isbell, C. J., Creasy, R. M., & Sherlock, J. W.
2009. Calibration and validation of the relative differenced Normalized Burn Ratio (RdNBR) to three
measures of fire severity in the Sierra Nevada and Klamath Mountains, California, USA. Remote
Sensing of Environment, 113(3), 645-656.
Ministry of Forests, Lands, Natural Resource Operations and Rural Development. 2014.
Biogeoclimatic zones of British Columbia. Government of British Columbia.
Moreno, Á., García-Haro, F. J., Martínez, B., & Gilabert, M. A. 2014. Noise reduction and gap filling of
fAPAR time series using an adapted local regression filter. Remote Sensing, 6(9), 8238-8260.
Munoz-Salinas, E., Rodriguez-Gonzalez, P. M., Casal-Roscón, L. M., & Ramirez-Suarez, A. 2018.
Post-fire vegetation recovery assessment using multi-temporal Landsat imagery in northwestern
Mexico. Journal of Applied Remote Sensing, 12(2), 155-171.
Natural Resources Canada. 2017. Canada's Ecozones and Ecoregions.
https://www.nrcan.gc.ca/maps-tools-publications/maps/canadas-ecozones-ecoregions/17050
Natural Resources Canada. (n.d.). Fire Weather Index (FWI). Canadian Wildland Fire Information
System (CWFIS). Retrieved from https://cwfis.cfs.nrcan.gc.ca/background/summary/fwiNational
Oceanic and Atmospheric Administration. 2021. ABI. GOES-R Series. Retrieved from
https://www.goes-r.gov/spacesegment/abi.html
Nelson Jr, R. M. 2001. Water relations of forest fuels. In Forest fires (pp. 79-149). Academic Press.
Nolan, R. H., Boer, M. M., Resco de Dios, V., Caccamo, G., & Bradstock, R. A. 2016. Large‐scale,
dynamic transformations in fuel moisture drive wildfire activity across southeastern Australia.
Geophysical Research Letters, 43(9), 4229-4238.
Norum, R. A. 1984. Measuring fuel moisture content in Alaska: standard methods and procedures
(Vol. 171). US Department of Agriculture, Forest Service, Pacific Northwest Forest and Range
Experiment Station.
Paltridge, G. W., & Barber, J. 1988. Monitoring grassland dryness and fire potential in Australia with
NOAA/AVHRR data. Remote sensing of Environment, 25(3), 381-394.
Parks, S.A., Dillon, G.K. and Miller, C., 2014. A new metric for quantifying burn severity: the
relativized burn ratio. Remote Sensing, 6(3), pp.1827-1844.
Pompe, A., & Vines, R. G. 1966. The influence of moisture on the combustion of leaves. Australian
Forestry, 30(3), 231-241.
Qi, Y., Dennison, P. E., Jolly, W. M., Kropp, R. C., & Brewer, S. C. 2014. Spectroscopic analysis of
seasonal changes in live fuel moisture content and leaf dry mass. Remote Sensing of Environment,
150, 198-206.
Quintano, C., Fernández-Manso, A., & Fernández-Manso, O. 2018. Combination of Landsat and
Sentinel-2 MSI data for initial assessment of burn severity. International journal of applied earth
observation and geoinformation, 64, 221-225.
Rao, K., Williams, A. P., Flefil, J. F., & Konings, A. G. 2020. SAR-enhanced mapping of live fuel
moisture content. Remote Sensing of Environment, 245, 111797.
Ryan, K. C. 2002. Dynamic interactions between forest structure and fire behavior in boreal
ecosystems. Silva Fennica, 36(1), 13-39.

115
Schneider, P., Roberts, D.A. and Kyriakidis, P.C., 2008. A VARI-based relative greenness from
MODIS data for computing the Fire Potential Index. Remote Sensing of Environment, 112(3),
pp.1151-1167.
Scott, J. H., & Burgan, R. E. 2005. Standard fire behavior fuel models: A comprehensive set for use
with Rothermel's surface fire spread model. USDA Forest Service, Rocky Mountain Research Station,
General Technical Report RMRS-GTR-153
Service, B. C. W. 2022. Wildfire season summary. Province of British Columbia. Retrieved January
27, 2023, from https://www2.gov.bc.ca/gov/content/safety/wildfire-status/about-bcws/wildfirehistory/wildfire-season-summary
Stocks, B. J., Fosberg, M. A., Lynham, T. J., Mearns, L., Wotton, B. M., & Yang, Q. 2004. Climate
change and forest fire potential in Russian and Canadian boreal forests. Climatic Change, 67(3), 233249.
Trowbridge, R., & Feller, M. C. 1988. Relationships between the moisture content of fine woody fuels
in lodgepole pine slash and the Fine Fuel Moisture Code of the Canadian Forest Fire Weather Index
System. Canadian Journal of Forest Research, 18(1), 128-131.
Tucker, C. J. 1979. Red and photographic infrared linear combinations for monitoring vegetation.
Remote sensing of Environment, 8(2), 127-150.
Tymstra, C., Stocks, B. J., Cai, X., & Flannigan, M. D. 2020. Wildfire management in Canada: Review,
challenges and opportunities. Progress in Disaster Science, 5, 100045.
USGS EROS Archive - Sentinel-2 - Comparison of Sentinel-2 and Landsat | U.S. Geological Survey.
2022. Retrieved from https://www.usgs.gov/centers/eros/science/usgs-eros-archive-sentinel-2
Van Wagner, C. E., & Forest, P. 1987. Development and structure of the Canadian forest fire weather
index system. In Can. For. Serv., Forestry Tech. Rep.
Van Wagner, CE, and TL Pickett. 1085. Equations and FORTRAN program for the Canadian forest
fire weather index system. Flight. 33. 1985.
Viegas, D. X., Viegas, M. T. S. P., & Ferreira, A. D. 1992. Moisture content of fine forest fuels and fire
occurrence in central Portugal. International Journal of Wildland Fire, 2(2), 69-86.
Wang, L., & Qu, J. J. 2007. NMDI: A normalized multi‐band drought index for monitoring soil and
vegetation moisture with satellite remote sensing. Geophysical Research Letters, 34(20).
Wang, X., Wotton, B. M., Cantin, A. S., Parisien, M. A., Anderson, K., Moore, B., & Flannigan, M. D.
2017. cffdrs: an R package for the Canadian forest fire danger rating system. Ecological Processes,
6(1), 1-11.
Wang, Y., Wei, X., del Campo, A. D., Winkler, R., Wu, J., Li, Q., & Liu, W. 2019. Juvenile thinning can
effectively mitigate the effects of drought on tree growth and water consumption in a young Pinus
contorta stand in the interior of British Columbia, Canada. Forest Ecology and Management, 454,
117667.
White, J. D., Ryan, K. C., Key, C. C., & Running, S. W. 1996. Remote sensing of forest fire severity
and vegetation recovery. International Journal of Wildland Fire, 6(3), 125-136
Williams Lake Community Forest. 2022. Retrieved January 27, 2023, from
https://williamslakecommunityforest.com/operations/

116
Wilson, E. H., & Sader, S. A. 2002. Detection of forest harvest type using multiple dates of Landsat
TM imagery. Remote Sensing of Environment, 80(3), 385-396.
Wotton, B. M. 2009. Interpreting and using outputs from the Canadian Forest Fire Danger Rating
System in research applications. Environmental and Ecological Statistics, 16(2), 107-131.
Xiong, Q., Wang, Y., Liu, D., Ye, S., Du, Z., Liu, W., Huang, J., Su, W., Zhu, D., Yao, X. and Zhang,
X., 2020. A cloud detection approach based on hybrid multispectral features with dynamic thresholds
for GF-1 remote sensing images. Remote Sensing, 12(3), p.450.
Yang, Z., Dan, T., & Yang, Y. 2018. Multi-temporal remote sensing image registration using deep
convolutional features. Ieee Access, 6, 38544-38555.
Yebra, M., Dennison, P. E., Chuvieco, E., Riano, D., Zylstra, P., Hunt Jr, E. R., ... & Jurdao, S. 2013.
A global review of remote sensing of live fuel moisture content for fire danger assessment: Moving
towards operational products. Remote Sensing of Environment, 136, 455-468.
Yoder, B. J., Ryan, M. G., Waring, R. H., Schoettle, A. W., & Kaufmann, M. R. 1994. Evidence of
reduced photosynthetic rates in old trees. Forest Science, 40(3), 513-5270.
Zald, H. S., & Dunn, C. J. 2018. Severe fire weather and intensive forest management increase fire
severity in a multi‐ownership landscape. Ecological Applications, 28(4), 1068-1080.
Zhang, D., & Zhou, G. 2016. Estimation of soil moisture from optical and thermal remote sensing: A
review. Sensors, 16(8), 1308.
Zhu, X., Chen, J., & Gao, F. 2019. Combining Landsat and Sentinel-2 data to improve forest species
classification in the northeastern United States. Remote Sensing of Environment, 231, 111228.
Zhou, X., Zhu, X., Dong, Z., & Guo, W. 2016. Estimation of biomass in wheat using random forest
regression algorithm and remote sensing data. The Crop Journal, 4(3), 212-219.

Appendix
Appendix 1: Sample sites around Prince George and Smithers
Site Name

Elevation (m) Latitude

Longitude

North Fraser Juvenile

710

54.24368

-122.48823

North Fraser Mature

698

54.242285 -122.48995

North Fraser Open

698

54.243852 -122.49074

Stone Creek Juvenile

860

53.6139

Stone Creek Mature

895

53.616179 -122.53591

Stone Creek Open

881

53.616222 -122.5379

-122.53723

117
Tamarac Juvenile

858

53.8701

-123.37023

Tamarac Mature

869

53.872643 -123.36881

Tamarac Open

865

53.871334 -123.37044

Barren Juvenile

1051

54.51044

Barren Mature

1043

54.511318 -126.61125

Barren Open

1059

54.50796

-126.61509

Chapman Juvenile

848

54.8839

-126.62353

Chapman Mature

854

54.882793 -126.6376

Chapman Open

809

54.88272

McDonnell Juvenile

990

54.780862 -127.52063

McDonnell Mature

981

54.781871 -127.50614

McDonnell Open

980

54.78217

Dennis Juvenile

883

54.761853 -127.46593

Dennis Mature

885

54.762177 -127.46537

Dennis Open

885

54.761427 -127.46753

-126.61434

-126.62893

-127.51434

Appendix 2: Sample sites around Prince George and Smithers
Fire Weather
Elevation (m) Latitude Longitude Sample site name Distance (km)
Station (FWS)
Bear Lake

715

54.482 -122.683

North Fraser

27.3

Hixon

615

53.411 -122.596

Stone Creek

22.1

Bednesti

825

53.865 -123.323

Tamarac

3.21

Houston

608

54.394 -126.618

Barren

11

Upper Fulton

900

55.034

Chapman

20.1

Pine Creek

1320

54.684 -127.326 McDonnell/Dennis

-126.8

18.5/12.5

118

Appendix 3: List of weather stations installed and used in this research in 2021-2022,
where Temp=Air temperature (°C), RH= Relative humidity (%), WS=Wind speed (Kph),
P=precipitation (mm), NF =North Fraser, SC= Stone Creek, TM=Tamarac,
MD=McDonnell, DS=Dennis, Br= Barren, and CP=Chapman. Notice all the tipping
buckets and wind sensors at juvenile and mature stands were only installed after midsummer of 2022. The recoding interval for 30 cm HOBO loggers were set at 15 mins
and one hour for open weather stations.
Site Name

Sensor Installed

Parameters
measured
Latitude

Hobo, wind
sensor, tipping
NF Juvenile bucket

TEMP, RH,
WS, P

NF Mature

Hobo, wind
sensor, tipping
bucket

TEMP, RH,
WS, P

NF Open

Hobo, weather
station, tipping
bucket

TEMP, RH,
WS, P

Hobo, wind
sensor, tipping
SC Juvenile bucket

TEMP, RH,
WS, P

SC Mature

Hobo, wind
sensor, tipping
bucket

TEMP, RH,
WS, P

SC Open

Hobo, weather
station, tipping
bucket

TEMP, RH,
WS, P

Hobo, wind
sensor, tipping
TM Juvenile bucket

TEMP, RH,
WS, P

Longitude

Elevation(m)

Date of data
collection
2021-06-08 to
2022-10-18

54.243689 -122.48823 710
2021-06-08 to
2022-10-18
54.242285 -122.48995 698
2021-06-08 to
2022-10-18
54.243852 -122.49074 698
2021-06-09 to
2022-10-18
53.613937 -122.53723 860
2021-06-09 to
2022-10-18
53.616179 -122.53591 895
2021-06-09 to
2022-10-18
53.616222 -122.53790 881
2021-06-09 to
2022-10-18
53.870100 -122.37023 858

TM Mature

Hobo, wind
sensor, tipping
bucket

TEMP, RH,
WS, P

53.872643 -123.36881 869

TM Open

Hobo, weather
station

TEMP, RH,
WS, P

53.871334 -123.37044 865

2021-06-09 to
2022-10-18

MD Juvenile Hobo

TEMP, RH

54.780862 -127.52063 990

2021-07-08 to 0903

MD Mature

TEMP, RH

54.781871 -127.50614 981

2021-07-08 to 0903

Hobo

2021-06-09 to
2022-10-18

119
Hobo, weather
station

TEMP, RH,
WS, P

54.782170 -127.51434 980

2021-07-08 to 0903

Hobo, wind
sensor, tipping
DS Juvenile bucket

TEMP, RH,
WS, P

54.761853 -127.46593 883

2022-07-06 to 1027

DS Mature

Hobo, wind
sensor, tipping
bucket

TEMP, RH,
WS, P

54.762177 -127.46537 885

2022-07-06 to 1027

DS Open

Hobo, weather
station

TEMP, RH,
WS, P

54.761427 -127.46753 885

2022-07-06 to 1027

Hobo, wind
sensor, tipping
BR Juvenile bucket

TEMP, RH,
WS, P

54.510444 -126.61434 1051

2021-07-09 to
2022-10-27

BR Mature

Hobo, wind
sensor, tipping
bucket

TEMP, RH,
WS, P

54.511318 -126.61125 1043

2021-07-09 to
2022-10-27

BR Open

Hobo, weather
station

TEMP, RH,
WS, P

54.507960 -126.61509 1059

2021-07-09 to
2022-10-27

Hobo, wind
sensor, tipping
CP Juvenile bucket

TEMP, RH,
WS, P

54.883900 -126.62353 848

2021-07-09 to
2022-10-27

CP Mature

Hobo, wind
sensor, tipping
bucket

TEMP, RH,
WS, P

54.882793 -126.63760 854

2021-07-09 to
2022-10-27

CP Open

Hobo, weather
station

TEMP, RH,
WS, P

54.882720 -126.62893 809

2021-07-09 to
2022-10-27

MD Open

Appendix 4: List of weather stations installed and used in this research in 2021-2022,
where T=Air temperature (°C), RH= Relative humidity (%), Wind=Wind speed (Kph),
Pre=precipitation (mm), for data precision, the number means precision after 0.

Sensor Name
Open weather
station
Hobo U23
sensors
Tipping bucket
Wind speed
sensor

Data
measured

Sampling
Frequency

T, RH,
Wind, Pre

Hourly

Data precision
T (0.001°C), RH/Pre
(0.1%/0.1mm), Wind
(0.01kph)

Data
measured
height (m)
2
0.3

T, RH
Pre

15 mins
hourly

T (0.001°C), RH(0.001%)
Pre (0.01%)

wind

15 mins

Wind (0.01mm)

0.3
2

120
Appendix 5: Average and variance of temperature (T, °C), relative humidity (RH, %),
wind speed (ws, kph) as well as Precipitation (P, mm) in different stands from data
collected in summer of 2022.
Stand

Avg T

Juvenile
Mature
Open

14.17
13.95
15.29

Avg
RH
10.49 72.71
11.94 79.64
14.2 72.61

Var T

Var
Avg
Var
Avg P Var P
RH
ws
ws
615.64
0.1
0.03
1.57 18.23
157.15
0.03
0
1.27 36.62
130.32
0.79
0.4
1.54 22.89

121
A

B

Appendix 6: A) Cloud and shadow-free true colour image, and B) NDVI for the North
Fraser site, image captured on June 2nd, 2022, centred at 54.252924 N, 122.499171 W.

122
A

B

Appendix 7: -A Cloudy and cloud-shadowed true colour image, and B) NDVI for the
North Fraser site, image captured on June 4th, 2022, centred at 54.252924 N,
122.499171 W.

123
A

B

Appendix 8: A) Cloud true colour image, and B) NDVI for the North Fraser site, image
captured on June 9th, 2022, centred at 54.252924 N, 122.499171 W.

124

Appendix 9: Daily averaged FMC in 2021 for different stands and fuel types at the
Prince George and Smithers study locations. The number of observations averaged to
create each point varies between 3 and 15, depending on the number of sites visited.