ArticlePDF Available

Abstract and Figures

The structure and function of the boreal forest are significantly influenced by forest fires. The ignition and growth of fires depend quite strongly on weather; thus, climate change can be expected to have a considerable impact on forest fire activity and hence the structure of the boreal forest. Forest fire occurrence is an extremely important element of fire activity as it defines the load on suppression resources a fire management agency will face. We used two general circulation models (GCMs) to develop projections of future fire occurrence across Canada. While fire numbers are projected to increase across all forested regions studied, the relative increase in number of fires varies regionally. Overall across Canada, our results from the Canadian Climate Centre GCM scenarios suggest an increase in fire occurrence of 25% by 2030 and 75% by the end of the 21st century. Results projected from fire climate scenarios derived from the Hadley Centre GCM suggest fire occurrence will increase by 140% by the
Content may be subject to copyright.
CSIRO PUBLISHING International Journal of Wildland Fire 2010,19, 253–271
Forest fire occurrence and climate change in Canada
B. M. WottonA,D,C. A. NockBand M. D. FlanniganC
ACanadian Forest Service–Natural Resources Canada, Faculty of Forestry, University of Toronto,
Toronto, ON, M5S 3B3, Canada.
BInstitute of Botany, University of Natural Resources and Applied Life Sciences Vienna,
1180 Vienna, Austria.
CCanadian Forest Service–Natural Resources Canada, Sault Ste. Marie, ON, P6A 2E5, Canada.
DCorresponding author. Email:
Abstract. The structure and function of the boreal forest are significantly influenced by forest fires. The ignition and
growth of fires depend quite strongly on weather; thus, climate change can be expected to have a considerable impact
on forest fire activity and hence the structure of the boreal forest. Forest fire occurrence is an extremely important
element of fire activity as it defines the load on suppression resources a fire management agency will face. We used two
general circulation models (GCMs) to develop projections of future fire occurrence across Canada. While fire numbers are
projected to increase across all forested regions studied, the relative increase in number of fires varies regionally. Overall
across Canada, our results from the Canadian Climate Centre GCM scenarios suggest an increase in fire occurrence of
25% by 2030 and 75% by the end of the 21st century. Results projected from fire climate scenarios derived from the
Hadley Centre GCM suggest fire occurrence will increase by 140% by the end of this century. These general increases
in fire occurrence across Canada agree with other regional and national studies of the impacts of climate change on fire
activity. Thus, in the absence of large changes to current climatic trends, significant fire regime induced changes in the
boreal forest ecosystem are likely.
Fire, which acts as a major stand renewing agent, has been an
integral part of life cycles of Canada boreal forest for thou-
sands of years. Fire influences both forest structure and function
(Weber and Stocks 1998). Fire activity is strongly influenced
by several factors: weather and climate, fuels, ignition agents
and human activity (Countryman 1972; Johnson 1992; Swetnam
1993). Today, both lightning and human-caused fires ignite and
spread across Canada’s forested landscape burning on aver-
age 2.5M ha annually and current statistics (based on the last
35 years) reveal that throughout most of the country, human- and
lightning-caused fires occur in roughly equal numbers (Stocks
et al. 2002). However, examination of the records of area burned
by fire typically reveals that lightning fires are the dominant
cause of the majority (80%) of the area burned in Canada due
to their potential for ignition in temporal clusters and in areas
far from human population (Stocks et al. 2002). These factors
make fire suppression more difficult for two reasons. First, a
large number of simultaneous ignitions can overwhelm a sup-
pression organizations initial attack capacity. Second, fires that
are located far from human population can typically evade detec-
tion for longer and when reported, can take longer to receive
suppression action due to the travel time needed for suppression
resources (Martell and Sun 2008).
Forest fire management in Canada is carried out by provincial
and territorial agencies as well as Parks Canada (who manage fire
in Canada’s National Parks system). Despite the independence
of these organizations, fire management activities occur in a
similar manner. Fires ignited in areas where fire policies dictate
they should be excluded (e.g. high value areas such as those
close to human settlements or forestry operations) are aggres-
sively attacked upon detection (called initial attack) with rapidly
deployed resources. In the small percentage of cases where this
initial attack is unsuccessful, resulting ‘escaped fires’ are gen-
erally managed for perimeter containment and extinguishment
with a larger, extended mobilization of resources.The success or
failure of this initial attack process has traditionally been one of
the main measures of the overall success or failure of a fire man-
agement agency at achieving their goals. Large spreading fires
in the boreal forest, which tend to spread as crown fires, output
large amounts of energy and are difficult and often impossible
to control through direct attack. The success of a fire agency in
terms of limiting area burned in high value areas thus depends on
finding and suppressing fires in those areas before they become
large and reach a steady-state of high energy output. It is these
escaped fires that grow large and lead to most of the area burned.
Across Canada, one finds that 97% of area burned is caused by
<3% of fires (Stocks et al. 2002). As such, part of an agency’s
daily operational activities are anticipating both the number of
fires expected as well as their location, and planning resource
alert levels and deployments to meet these needs.
Daily preparedness planning by the fire management agen-
cies in Canada relies upon the Canadian Forest Fire Danger
Rating System (CFFDRS) to provide an understanding of fire
potential on the landscape given the weather systems affecting
the region and the forest fuel types in a management area. The
CFFDRS (described by Stocks et al. 1989) contains two major
sub-systems that assist in these daily fire management planning
© IAWF 2010 10.1071/WF09002 1049-8001/10/030253
254 Int. J. Wildland Fire B. M. Wotton et al.
decisions: a system that predicts the behavior of fire in vari-
ous fuel types (Forestry Canada Fire Danger Group 1992), and
a system that provides three indicators of moisture in different
layers of the forest floor as well as several relative indicators
of potential fire behavior (Van Wagner 1987). The outputs of
this latter system, called the Canadian Forest FireWeather Index
(FWI) System, are determined based on daily observation of fire
weather conditionsAat weather stations across an agency’s fire
management area and are used to assist fire managers in estimat-
ing expected fire occurrence, potential fire spread and intensity.
The FWI System tracks moisture in three distinct fuel layers.
The moisture content of surface litter is characterized by the Fine
Fuel Moisture Content (FFMC) and is important to determin-
ing the sustainability and vigor of surface fire spread (Forestry
Canada Fire Danger Group 1992; Lawson and Armitage 1997;
Beverly and Wotton 2007). The moisture content of the upper
portion of the organic layer in the forest floor is tracked by the
Duff Moisture Code (DMC) and is important to sustainability
of smouldering and fuel consumption in the forest floor (Van
Wagner 1972; Frandsen 1987). The moisture content in deeper
organic layers or in large pieces of woody debris on the forest
floor is indicated by the Drought Code (DC) and is a useful
indicator of extreme dryness and drought conditions that have
the potential to make fire suppression more difficult and time-
consuming. The FFMC, DMC and DC are each influenced by
weather conditions to different extents. All are strongly influ-
enced by temperature and rainfall amount; however, the FFMC
is also influenced by relative humidity and wind and the DMC
is additionally affected by relative humidity.
There has been significant research into the factors influenc-
ing the day-to-day variation in the numbers of fires occurring
in boreal regions. These studies have shown one of the fac-
tors most strongly influencing variation in the expected number
of human-caused fires is moisture content of the surface lit-
ter layer (represented by the FFMC) (e.g. Martell et al. 1987,
1989; Poulin-Costello 1993; Vega-Garcia et al. 1995; Wotton
et al. 2003). Further studies have shown that the probability of
lightning fire ignition is linked to the amount of moisture in the
organic soils in the upper part of the forest floor (those soils
represented by the DMC) (Flannigan and Wotton 1991; Wotton
and Martell 2005). Other studies have shown that factors such as
forest type also influence the number of lightning-caused fires
(Krawchuk et al. 2006) and human influences, such as the loca-
tion of population centres, roads and campgrounds also influence
the absolute numbers of expected human-caused fires in a region
(e.g. Vega-Garcia et al. 1995; Pew and Larsen 2001). However,
when fire management agencies are trying to make predictions
of the daily number of fires expected in a particular region, they
tend to ignore fine scale variation in these static elements and
focus on the dynamic changes in the moisture content of fuels
which in turn have been affected by the movement of weather
systems through their area of interest.
Climate change is a well recognized critical issue threatening
to (and has, in many opinions, already begun to) put significant
pressure on societies and ecosystems around the world. Canada
is particularly sensitive to impacts of climate change due to its
extensive high latitude ecosystems where changes are expected
ATemperature, relative humidity, 10 m open wind speed and 24-h accumulation of rainfall all recorded at 1200 hours Local Standard Time (LST).
to be significant (i.e. the arctic and boreal regions). Such impacts
have been widely studied (e.g. Soja et al. 2006; Anisimov et al.
2007; Field et al. 2007). General circulation model (GCM) sce-
narios show expected temperature increases across Canada in the
order of 3–5C by the end of the 21st century (Lempriere et al.
2008). Overall, warmer temperatures can be expected to increase
evapotranspiration, lower watertables (Roulet et al. 1992) and
decrease surface soil or fuel moisture content unless there are
significant increases in precipitation. GCMs projections indicate
that summer precipitation amounts and patterns across Canada
are expected to change as well, with some areas seeing increased
rainfall while other seeing a decrease (Flannigan et al. 2000). It is
important to understand, however, that there is much lower con-
fidence in GCM projections of precipitation than in projections
of temperature.
Because fire activity is strongly linked with weather and
fire plays such a major role in the life cycle of Canada’s for-
est, research into the potential impacts of climate change on
fire activity in Canada has been ongoing for some time (Flanni-
gan and Van Wagner 1991; Stocks 1993; Wotton and Flannigan
1993; Bergeron and Flannigan 1995; Flannigan et al. 1998, 2000;
Stocks et al. 1998). This early work focussed on the impacts
of climate change on overall fire season severity using output
from multiple GCMs and later, scenarios generated by Regional
Climate Models (RCMs) showed that while there were some
strong regional differences in Canada, overall fire seasons would
increase both in length and in severity.
More recently, such climate change impacts research has
focussed on predicting more physically basic (and easily under-
standable by the public and policy-makers) characteristics of
forest fire activity such as numbers of fires and area burned.
Flannigan et al. (2005) used correlations developed between
historic area burned and various elements of fire weather (most
commonly temperature) to develop projections of future area
burned for the various ecozones of Canada. These results, which
used the same GCM scenarios as in the current study, showed that
area burned could be expected to double (from current averages)
by the end of the 21st century. In a study specifically examin-
ing fire growth in the province of Alberta, Tymstra et al. (2007)
simulated growth of fires on the landscape using weather from
RCM scenarios and found that in terms of fire growth potential
alone, area burned increased 30% by the end of the current cen-
tury. Recently, using future GCM climates generated for western
boreal North America, Balshi et al. (2008) projected even larger
increases in area burned in the future: 3.5 to 5.5 times current
averages by the end of the 21st century.
Drever et al. (2008) studied large fire occurrence in the decid-
uous forests of western Quebec (Témiscamingue region) and
suggested some increases in area burned were possible in this
ecosystem. In a study of a large forested area in the province of
Alberta, Krawchuk et al. (2009) used RCM scenarios along with
previously developed relationships linking climate, fire danger
and forest type to fire occurrence (Krawchuk et al. 2006) and
projected an 80% increase in lightning fire occurrence by the end
of the century. Wotton et al. (2003) studied the influence of cli-
mate change on human-caused fire occurrence in the province of
Ontario using the daily ecoregion-based fire occurrence models
Fire occurrence and climate change Int. J. Wildland Fire 255
Fig. 1. Ecoregion-based spatial units from across Canada (shaded) used to summarize weather and fire activity and develop fire occurrence models.These
are based on the national framework of ecoregions of Canada and provincial political and fire management zone boundaries.
they developed.They found potential increases of 50% by the end
of the 21st century. In a follow-up study, Wotton et al. (2005)
generated scenarios of both human- and lightning-caused fire
occurrence and used these with Ontario’s level of protection
analysis system (LEOPARDS; McAlpine and Hirsch 1999) to
show that overall fire occurrence could be expected to increase
by 50% by the end of the 21st century while the number of fires
that escape initial attack is expected to be 80% higher than
current levels. A detailed simulation study carried out byWotton
and Stocks (2006) using LEOPARDS examined the impact of
changing resource levels on managing increased fire activity.
Future scenarios used in that study were quite conservative and
indicated only a 15% increase in the number of fires by the year
2040. However, because increases in the number of fires push fire
management agencies to the limits of their capacity more often,
in that simulation, Ontario’s fire management agency needed
to more than double current resource capacity (i.e. fire crews,
helicopters, airtankers) to maintain present levels of escape fires.
In this research, we expanded studies of forest fire occurrence
carried out in Ontario to examine the impacts of climate change
on fire occurrence in the rest of the forested ecoregions of Canada
that have had significant forest fire activity over the past 25 years
(Fig. 1). We use a previously developed and tested generalized
linear model structure for fire occurrence prediction and develop
simple models that allow us to predict daily fire occurrence
in individual forested ecoregions across the country based on
fire weather and fuel moisture information. These models of the
daily expected number of fires in each ecoregion are then used
with GCM scenarios to assess potential changes in annual fire
occurrence with climate change. This climate change impact
study focuses on provinces and territories with significant areas
of forest under fire management protection. Due to unavail-
able information on fire report and fire weather, this study does
not include the Atlantic provinces in eastern Canada and the
unforested portions of the prairie provinces and territories.
Datasets obtained from individual provincial fire management
agencies varied somewhat in their temporal extent as well as
their breakdown of fire causes into subcategories. In some cases,
256 Int. J. Wildland Fire B. M. Wotton et al.
Table 1. Range of years of fire occurrence and fire weather data used
in the current analysis for each province and territory of Canada
Province or territory Years of fire and Total number Total area
weather data of fires bur ned (ha)
British Columbia 1980–1999 49 291 1296 500
Alberta 1983–2004 23 697 2420 000
Saskatchewan 1981–2005 17 697 9434 500
Manitoba 1980–1999 11 446 9094 700
Ontario 1980–2005 37 988 5296 600
Quebec 1985–2000 14 960 6070 500
Yukon 1980–1999 3760 3097 900
Northwest Territories 1980–1999 6696 13 742 300
the length of the fire weather datasets did not match fire datasets.
Where these discrepancies occurred, only the subset of the data
where all information was available was used. Table 1 shows the
range in years for each provincial or territorial dataset used in this
analysis. While some province’s digitized fire records went back
many decades, we used data only from 1980 onward. We chose
this cut-off to our datasets for two main reasons: (1) in modelling
human-caused fires, changes in human use of the forest will take
place over extended periods of time; and (2) before the advent
of lightning detection systems (around 1980 in most provinces),
the distinction between lightning- and human-caused fires was
likely to be subject to error (Stocks et al. 2002).
Fire data
While fire management agencies in Canada operate in similar
ways, the format and specific details of their records vary con-
siderably. In general however, details on the ignition, spread and
extinguishment of all fires detected and reported to the agency
are archived as part of standard operational procedures. In recent
decades, this data has been stored in digital form by each agency.
This data takes the form (within every fire agency in Canada) of
individual point records for each fire that occurs. Fires are spa-
tially referenced with the latitude and longitude of the known
or estimated ignition point and other attributes about the fire
are based on observations by the fire management organization.
For this analysis, we obtained only a subset of these agency fire
records: general cause type, location (latitude, longitude), start
date, report date and final area burned.
Weather station data
Each fire management agency operates their own fire weather
observing network recording daily observations (at 1200 hours
Local Standard Time) to estimate the fuel moisture codes and
fire behavior indices of the Fire Weather Index (FWI) System.
This information is in turn used for pre-suppression planning
and resource deployment. Environment Canada, a federal agency
responsible for monitoring and forecasting environmental con-
ditions throughout the country, also operates weather stations
throughout the country and some fire management agencies
augment their own weather station networks’ observations with
observations from the Environment Canada network. We used
weather station records both from provincial networks and
Environment Canada stations (if fire weather records were not
available digitally in a province for a specific period). Data
obtained from these records were: temperature, relative humid-
ity, wind speed, rainfall, FFMC, DMC, DC. Data from three other
outputs of the FWI System were also obtained: Initial Spread
Index (ISI), Build-Up Index (BUI) and FWI.
While some agencies have more than 200 stations in their
networks, the large spatial extent of the Canadian boreal forest
means there are gaps in coverage. Agencies typically spatially
interpolate fuel moisture and fire behavior indices to estimate fire
danger conditions between fire weather station locations. Flan-
nigan and Wotton (1989) provide an evaluation of commonly
used interpolation techniques in fire management in Canada
and describe the thin-plate cubic spline method we used in this
current study.
Future fire weather scenarios
Numerous GCMs have been developed by different groups
around the world and several these scenarios have been selected
by the Intergovernmental Panel on Climate Change (IPCC) for
their recent assessments of global climate change impacts (IPCC
2001, 2007). Numerous model intercomparisons have been car-
ried out (Covey et al. 2003; Meehl et al. 2007; Randall et al.
2007). For the development of future scenarios of forest fuel
moisture that can be used to realistically drive models of daily
expected numbers of fires, daily GCM output is needed across
the area being studied. Using these extremely large datasets, data
reduction and the development of daily fire weather datasets
that are useful inputs to existing fuel moisture models (e.g.
Van Wagner 1987) can be quite time consuming and com-
putationally complex. In the process of data reduction and
interpretation, it can also be helpful to have direct relationships
with the GCM modellers responsible for original scenario devel-
opment to understand assumptions and limitations in the model.
We obtained our daily datasets from two well established GCMs:
the Canadian Climate Centre (CCC) and from the UK’s Hadley
Centre (HAD). These GCMs have both been used by the inter-
national climate change impacts modelling community for well
over 15 years, both used in the last three IPCC assessment reports
(IPCC 1995, 2001, 2007) and provide projections consistent with
other global warming projections (Covey et al. 2003; Meehl
et al. 2007; Randall et al. 2007).
From the CCC, daily data was obtained from the first gen-
eration coupled ocean–atmosphere model (CGCM1; Flato et al.
2000) for three 21-year time slices spanning 1975–1995, 2020–
2040 (referred to hereafter as the 2030 scenario) and 2080–2100
(referred to hereafter as 2090 scenario). This model included
both greenhouse gas and sulfate aerosol forcing contributing to
a 1% increase in CO2per year. This is similar to an A2 emis-
sion scenario used in the fourth IPCC assessment (IPCC 2007),
a ‘business-as-usual’ scenario very commonly used in climate
change impacts studies. Given that greenhouse gases have been
found recently to be increasing at rates faster than this 1% per
year (Raupach et al. 2007), we believe that for the purposes of
this study, this represents a realistic ‘middle of the road’ emis-
sion scenario for the future. The time period 2080–2100 roughly
corresponds to an equivalent 3×CO2scenario when including
the net radiative effect of all the greenhouse gases. The grid
spacing is 3.75longitude by 3.75latitude. Daily data from
Fire occurrence and climate change Int. J. Wildland Fire 257
the Hadley Centre was obtained for time slices from 1975–1990
and the 2090 scenario from the HadCM3 model (Hulme et al.
1999). This implementation of the Hadley model, more formally
known as HadCM3GGa1, contained only greenhouse gas forc-
ing, and again output from the 2080–2099 time slice was roughly
equivalent to a 3×CO2scenario.The identical 21-year time peri-
ods from the CCC GCM could not be reproduced exactly using
the HadCM3 GCM results due to lack of availability of the
daily climate variables needed for the full time period. More
detail on these daily datasets and their use in the development of
future fire weather and fuel moisture scenarios are described in
detail in previous research (Logan et al.2004;Flanniganet al.
Daily fuel moisture observations were calculated for each
GCM grid cell across Canada from the daily GCM weather
streams and the FWI System for each of the future scenarios.
Outputs at the GCM grid cell level were not used directly in ana-
lysis but inter polated, using thin-plate cubic splines, to the centre
of each ecoregion to avoid any bias from using an individual grid
cell as representative of a specific point. These ecoregions and
the reasons for their use are described in the next section.
In previous studies of forest fire occurrence (e.g. Martell et al.
1987, 1989; Poulin-Costello 1993; Vega-Garcia et al. 1995), it
has been found that if occurrences are considered over small rel-
atively homogenous areas, expected daily fire occurrence can be
modelled reasonably well using the assumption that it follows a
Poisson distribution. We chose to use the ecoregions of Canada
as the basic spatial unit over which to summarize daily fire and
weather information, following the approach of Wotton et al.
(2003). These ecoregions, defined by the Ecological Stratifica-
tion Working Group (1996), are areas with relatively uniform
topography, underlying soil characteristics and forest species
composition and as such, provide ideal spatial units area over
which to summarize fire weather, fuel moisture and forest fire
occurrence. Because each province and territory in Canada car-
ried out fire management activities independently, ecoregions
that spanned political borders were further broken down along
provincial boundaries to eliminate any potential differences from
province to province. Within several provinces (Saskatchewan,
Manitoba, Ontario, Quebec), the area under fire management is
broken up into full suppression zones and observation zones. In
the former, policy states that all fires are actively suppressed. In
the latter, fires are generally monitored and suppressed only if a
fire threatens human values (e.g. lives and property in a northern
community or infrastructure supporting northern communities).
Where these fire management zones crossed an ecoregion bor-
der, we split the ecoregion along that fire management zone
border to maintain a relatively homogeneous level of fire man-
agement for each unit. A map of the ‘ecoregion’breakdown used
in this study is in Fig. 1. Throughout the remainder of this paper,
‘ecoregion’refers to these standard ecoregion units broken down
by political and fire management zone boundaries.
Human- and lightning-caused fires are different in their
ignition characteristics: the expected number of human-caused
ignitions in a region is strongly driven by moisture content of fine
surface fuels (Martell et al. 1987, 1989; Wotton et al. 2003),
while lightning-caused ignitions are most strongly influenced
by moisture in the organic layer where lightning ignitions can
smoulder and holdover (Anderson 2002; Wotton and Martell
2005). Because of these differences, expected numbers of
human- and lightning-caused fires are most effectively modelled
as two distinct processes. Furthermore, fire agencies typically
break down human-caused category into several major sub-
classifications which we assembled (where our data provided
enough information) into two distinct groups: those dominated
by a significant spring season, and those showing a mid-summer
peak in activity (as in Martell et al. 1987). Numbers of daily
fires in each cause group were summarized for each ecore-
gions shown in Fig. 1. Daily fire weather, fuel moisture and
fire behavior indices from the FWI System (both from the actual
and GCM-derived fire weather streams) were interpolated to the
approximate centroid of each ecoregion to provide a matching
daily weather record to the fire occurrence record for the region.
Human-caused fire occurrence modelling
Previous studies have shown that expected number of human-
caused fires occurring daily in a region can be reasonably
modelled with a Poisson distribution (Martell et al. 1987, 1989;
Poulin-Costello 1993; Mandallaz and Ye 1997; Wotton et al.
2003). In these models, moisture in the surface fuels is always
Fine Fuel Moisture Code (FFMC) output of the FWI System has
been shown to be a significant and strong predictor of expected
number of human-caused fires in regions of the boreal forest
(Martell et al. 1987, 1989; Wotton et al. 2003). The FFMC tracks
moisture in the litter fuels on the surface of the forest floor: the
fuels that most directly influence the ignition and spread of a
surface fire. Martell et al. (1987, 1989) also found an influence
of moisture content in heavier fuels through the significance of
the FWI System’s BUI, which is functionally a weighted mean of
the FWI System’s DMC and DC. Wotton et al. (2003) explicitly
examined DMC and DC and found they played a significant role
in determining expected number of fires occurrence in a region.
The relationships between fire occurrence and fuel moisture
(FFMC, DMC and DC) described in the preceding paragraph
were included in developing the models for this study using gen-
eralized linear modelling (see McCullagh and Nelder 1989 for
statistical background on generalized linear models). Using the
approach of these previous fire occurrence studies, we assumed
fires occurred as a Poisson process and thus chose a logarithmic
link function for the dependent variable (number of fires) and
the modelling assumption that residual errors would be from a
Poisson distribution.
Martell et al. (1989) showed that human-caused fires could
be grouped into two main sub-categories: those with a peak
in activity in the spring and those with a peak in activity
in the summer. Where our fire data included the distinction
between different categories of human-caused fires, we exam-
ined the seasonal distribution of fire cause-subtype (e.g. railway,
recreation) and classified each using a simple binary variable
(CAUSE_GROUP: 1 for spring and 0 for causes with a sum-
mer peak). Typically, fire cause types grouped into ‘spring peak’
category exhibited a strong peak in fire occurrence during the
month of May; fire cause types grouped into the ‘summer peak’
258 Int. J. Wildland Fire B. M. Wotton et al.
category, while some spring fire activity did occur, exhibited a
strong peak during the months of July and August.
Models were developed for each province separately because
of potential differences in record length and potential differ-
ences in the classification of fires into cause categories by each
agency. Wotton and Beverly (2007) characterized the relation-
ship between FFMC and actual litter moisture for a range of
forest types across Canada and showed that the relationship
can change with forest type as well as season in the year (i.e.
spring or summer). Indeed, Wotton et al. (2003) found that the
strength of the FFMC coefficient in their human-caused fire
models changed across ecoregions with significant different for-
est composition. To account for coarse scale change in forest
type across provinces, ecoregion was used as a categorical pre-
dictor in the model (labelled ECOREGION). In an attempt to
account for Wotton and Beverly’s (2007) observation that the
relationships between fuel moisture and FFMC can change with
forest type, an interaction term between ecoregion and FFMC
was also included. Additionally, because Wotton and Beverly
(2007) found that the relationship between the FFMC and actual
surface litter moisture can be different between spring and sum-
mer (likely a result of forest canopy closure), a simple binary
season variable (SEASON: 1 for days before 1 June; 0 for days
after 1 June) was also included in the model to attempt to account
for some of these differences.
A general model, using the key predictor variables identi-
fied from previous research (and described in the preceding
paragraphs) was then fit in each province using the model form:
+α7·SEASON (1)
where NHUM represents the total number of fires in an ecoregion
on a particular day.
Lightning-caused occurrence modelling
Lightning fire occurrence was well correlated with level of mois-
ture in the organic layer as indicated by the Duff Moisture
Code (DMC) of the FWI System (Flannigan and Wotton 1991;
Krawchuk et al. 2006) In fact, in the development of models for
daily prediction of lightning fire occurrence, Wotton and Martell
(2005) showed that DMC was an excellent indicator of the prob-
ability of having an ignition of the forest floor. Other lightning
fire prediction models developed in Canada, have also taken
advantage of this relationship (Kourtz andTodd 1992; Anderson
2002). Wotton (2009) described how these relationships hold
across Canada as a whole. Thus, DMC was included as a key
predictor in the current models. Wotton and Martell (2005) also
BThis classification of the current and previous day’s total rainfall (RT) into categories was based on a simple analysis of rainfall associated with lightning in
an ecoregion in central Alberta and Saskatchewan (ecoregion 147, the Mid-Boreal Uplands) that we believed was representative of the boreal forest, using
data from 1984–2004. The goal was to make a simpler classification system for rainfall because of the strongly skewed nature of the daily rainfall distribution.
The rainfall levels simply corresponded to the 25th, 50th, 75th, 90th and 95th percentiles of daily rainfall that occurred on days with lightning in this region.
This new variable, RCLASS, was thus classified as follows. RCLASS =0, RT 0.3; 1, 0.3 <RT0.9; 2, 0.9<RT2.3; 3, 2.3<RT4.8; 4, 4.8 <RT7.3;
5, RT>7.3. All values are in mm. Similar category breaks were found for neighboring ecoregions.
CModel results from B. M. Wotton (unpubl. data) from the implementation of the Wotton and Martell (2005) model into operations in the province of Ontario.
Similar results were also found for models developed for the province of Saskatchewan using the same model form.
found that the other moisture indictors of the FWI System had
an influence (albeit less than DMC) on probability of ignition
of a lightning fire. Thus, as in their models, FFMC and DC were
included in the model form used in this study.
In this study, we did not have lightning records for the full
country or period studied that would allow us to make use of the
same type of models developed by Wotton and Martell (2005).
Therefore, we model daily lightning fire occurrence on an ecore-
gion basis using the same approach as human-caused fires:
Poisson regression. We included a categorical classification of
daily rainfall amount as a surrogate for summertime lightning
activity because lightning occurrence rate and rainfall intensity
have been found to be correlated in numerous studies (e.g. Gosz
et al. 1995; Sheridan et al. 1997; Tapia et al. 1998; Anderson
2000). In this categorization, we examined rainfall occurring on
the current and previous day (to account for storms that had
occurred during the previous day) and classified the total rain-
fall on these 2 days into a five level categorical variable (called
RCLASS in the model in Eqn 2)B.
As with the human-caused fire occurrences, models were
developed for each province individually to avoid any potential
operational differences from province to province. In addition,
to account for potential differences in the relationships between
DMC and actual forest floor moisture between forest type
(Lawson et al. 1997; Wilmore 2001; Otway et al. 2007), ecore-
gion was included as a categorical predictor (ECOREGION)
along with an interaction between DMC and ecoregion. Indeed,
Wotton and Martell (2005) showed that the strength of DMCs
influence on probability of ignition varied between ecoregions.
Previous model development and operational implementation
of the model developed by Wotton and Martell (2005) found
that the inclusion of a binary season predictor (delineating the
spring and summer seasons) increased predictive power of the
relationshipsC; thus, a simple binary season indictor (SEASON:
1 for before 1 June; 0 for after 1 June) was also added to the
model, as was the case for the human-caused fire models. The
general model fit in each province, which followed from the
methods of Wotton and Martell (2005), had the form:
ln(NLTG)=β0+β1·ECOREGION +β2·DMC
+β5·DC +β6·SEASON +β7·RCLASS (2)
where NLTG represents the total number of fires in an ecoregion
on a particular day.
Future fire occurrence scenarios
Daily fire weather and fuel moisture values based on the future
GCM scenarios from Canadian Climate Centre and the Hadley
Centre were interpolated to the centroids of each ecoregion.
Fire occurrence and climate change Int. J. Wildland Fire 259
Annual number of fires
Fig. 2. Annual human-caused fire occurrence rates (number of fires per year) for ecoregions throughout the study
area. Numbers in the legend represent mean annual number of fires. The number of years each value is based on varies
from province to province (from 16 to 26 years) due to the length of fire records available from provincial fire agencies
(Table 1).
We examined how values of FFMC and DMC (two main predic-
tors of human- and lightning-caused fires respectively) changed
between the GCM scenarios by calculating changes in the
90th percentile levels of these distributions. Ninetieth percentile
values rather than means or medians were chosen because exam-
ining the tails of these moisture code distributions tends to be
more revealing of true levels of fire potential since the majority
fire activity occurs on these drier days (Flannigan and Wotton
Future fuel moisture scenarios from the GCMs were then
used with the models described in the previous two sub-sections
to generate future scenarios of fire occurrence. Daily fire occur-
rence predictions were aggregated to create estimates of annual
fire occurrence rates in each ecoregion, as well as each for each
DMC, which strongly influences expected lightning fire
occurrences, is an open ended index; that is, as the forest floor
dries it continues to increase in value with no true maximum.
After initial data analysis, we concluded that because of the expo-
nential link function in the Poisson model form, it would be
unrealistic to allow values of the DMC to exceed levels observed
currently in Canada. Therefore, we calculated a maximum DMC
value for each ecoregion from the historical dataset and did not
DHistorical DMC maximums were only exceeded in the Hadley 2090 general circulation model (GCM) scenario in the provinces of British Columbia, Ontario
and Quebec and only on 0.08, 0.6 and 2.3% of ecoregion-days respectively and in the CCC 2090 GCM only in Ontario on 1.8% of ecoregion-days.
allow DMCs in future GCM-based scenarios values to exceed
this value. For the very small number of daysDwhere a future
DMC value exceeded the observed present day maximum for
an ecoregion, its value was set equal to the present day maxi-
mum. While the introduction of this DMC cap was a conservative
assumption, we believed that it was reasonable given that it
reduced the potential for unrealistic predictions from the fire
occurrence model. FFMC, the main predictor of the human-
caused fire models and an important predictor in the lightning
fire models, has an upper limit built into the code and thus did
not require a similar limiting function.
Figs 2 and 3 show maps of present day average annual fire
occurrence for each of the ecoregions of our study area. The
differences in size of each ecoregion makes absolute compar-
isons of occurrence rates somewhat difficult (particularly as
large ecoregions tend towards more northern areas); however,
as would be expected, high human-caused fire occurrence rates
tend towards the high population density areas of the coun-
try (e.g. central and eastern Ontario, south-western Quebec,
central southern British Columbia). Overall, mean annual fire
occurrence rates in ecoregions vary from a minimum of two
260 Int. J. Wildland Fire B. M. Wotton et al.
Annual number of fires
Fig. 3. Annual lightning-caused fire occurrence rates (number of fires per year) for ecoregions throughout the study
area. Numbers in the legend represent mean annual number of fires. The number of years each value is based on varies
from province to province (from 16 to 26 years) due to the length of fire records available from provincial fire agencies
(Table 1).
Table 2. Human-caused fire occurrence rates for each province or territory in Canada using data observed from each fire agencies records and
expected fire occurrence generated using fire occurrence models and general circulation model (GCM) datasets
The range in observed years over which this actual estimate has been calculated is listed in Table 1. CCC, Canadian Climate Centre; HAD, Hadley Centre
Province or territory Average annual human-caused fire occurrence rate (percentage change from baseline GCM scenario)
Observed CCC 1975–1995 HAD 1975–1990 CCC 2020–2040 CCC 2080–2100 HAD 2080–2099
British Columbia 1185 565 606 618 (9) 683 (21) 1145 (89)
Alberta 463 234 196 256 (9) 277 (18) 303 (54)
Saskatchewan 359 208 73 218 (5) 234 (12) 114 (57)
Manitoba 277 335 135 430 (28) 535 (60) 227 (68)
Ontario 821 610 411 734 (20) 928 (52) 837 (104)
Quebec 644 305 269 339 (11) 423 (38) 632 (135)
Yukon 66 23 25 23 (0) 25 (10) 30 (20)
Northwest Territories 107 112 66 145 (29) 137 (22) 61 (8)
Total 3922 2392 1781 2763 (16) 3242 (26) 3349 (88)
human-caused fires per year in ecoregion 71 of Manitoba to a
maximum of 299 lightning-caused fires per year in the north-
ern half of ecoregion 205 (labelled 2051 in Fig. 1) of British
Columbia. Overall, annual occurrence rates in the provinces and
territories are in Tables 2 and 3.
Development and description of daily fire occurrence models
was not the primary purpose of this paper but was a neces-
sary first step to achieving the objective of examining potential
changes in forest fire occurrence rates across the country under
climate change scenarios. As such, we used established method-
ologies (following Wotton et al. 2003; Wotton and Martell 2005)
and examined known fire occurrence predictors to develop the
ecoregion-based models of fire occurrence in each province. We
have not attempted to explore and document new significant rela-
tionships between current environmental factors and fire occur-
rence in each province, but merely to form sound models of fire
occurrence based on the best current understanding of forest fire
occurrence and the datasets available. Thus, we will not present
Fire occurrence and climate change Int. J. Wildland Fire 261
Table 3. Lightning-caused fire occurrence rates for each province or territory in Canada, using data observed from each fire agencies records and
expected fire occurrence generated using fire occurrence models and general circulation model (GCM) datasets
The range in observed years over which this actual estimate has been calculated is listed in Table 1. CCC, Canadian Climate Centre; HAD, Hadley Centre
Province or territory Average annual lightning-caused fire occurrence rate (percentage change from baseline GCM scenario)
Observed CCC 1975–1995 HAD 1975–1990 CCC 2020–2040 CCC 2080–2100 HAD 2080–2099
British Columbia 1259 563 978 612 (9) 679 (21) 3409 (350)
Alberta 533 634 245 712 (12) 815 (29) 522 (110)
Saskatchewan 335 579 80 635 (10) 724 (25) 163 (100)
Manitoba 287 546 124 750 (37) 1096 (100) 257 (110)
Ontario 635 982 393 1596 (62) 3475 (250) 1202 (210)
Quebec 288 201 124 209 (4) 253 (26) 324 (160)
Yukon 86 22 18 24 (10) 30 (40) 24 (34)
Northwest Territories 209 222 77 375 (69) 345 (55) 78 (1)
Total 3632 3749 2039 4913 (31) 7417 (98) 5979 (190)
the results of this modelling in detail. Model forms are presented
in Appendices 1 and 2 (available as an Accessory publication,
available from the journal online) for those interested in using
the models in other climatological studies of fire occurrence. A
short description of some of the general model results follows.
Across the provinces, FFMC was indeed a strong predictor
of the expected daily number of human-caused fire arrivals in a
region. This agrees with the results of numerous studies (Martell
et al.1987,1989;Poulin-Costello1993;Vega-Garciaet al. 1995;
Wotton et al. 2003, 2005). There were significant differences
between ecoregions both in terms of absolute number of fires
predicted and through ecoregion interactions with the FFMC:
this agrees with the findings of Wotton et al. (2003) in Ontario.
The human-caused fire subtype grouping also had a strongly sig-
nificant influence on expected number of fires in all provinces
where data existed as did the seasonal breakdown (i.e. spring
and summer).
Our lightning-caused fire models across the country con-
firmed what has been found in other research done in Ontario
(Flannigan and Wotton 1991; Wotton and Martell 2005) and
Alberta (Krawchuk et al. 2006): DMC was a very good indictor
of lightning fire ignition. It was statistically significant in each
province or territories model. FFMC was also a significant fac-
tor influencing the number of lightning-caused fires expected
on any particular day, which was also found in previous research
from which the model used here were based (Wotton and Martell
2005). FFMC plays an important role in lightning fire occurrence
because of its influence on arrival probability; that is, the proba-
bility that a fire will be active enough (generating sufficient heat
and smoke) to be detected and reported to a fire management
Fig. 4 is a log-log plot of total number of fires predicted by the
models for each forested ecoregion across the country for each
year of the record compared with observed numbers. We plot
in log-log space because for most ecoregions, the predicted and
observed numbers of fires are relatively small. Therefore, there
is a high degree of clustering at less than 100 fires per year. In
addition, the non-transformed predicted v. observed plot shows
a characteristic increasing of variance with increasing fire num-
bers, which can be expected from a process that can be modelled
with a Poisson distribution. The plots show general agreement
between predicted and observed numbers for both human- and
lightning-caused fires. Correlation coefficients for the relation-
ship in transformed space were r2=0.87 (n=1668, P<0.0001)
and r2=0.76 (n=1672, P<0.0001) respectively; correlation
coefficients for the untransformed values were similar (r2=0.88
and 0.76 respectively). These levels of correlations were sim-
ilar to those found in a study of lightning caused fires in
Ontario (r2=0.86) by Wotton and Martell (2005). While the
relationships appear reasonable over a wide range of annual fire
occurrence, Fig. 4 does appear to indicate a slight over-prediction
in the models for low fire years, particularly for the lightning fire
Climate change scenarios
Fig. 5 shows summertime (May through August) temperature
and precipitation change across the country in the 2030 and
2090 scenario for the CCC GCM as summarized within each
ecoregion. The temperature maps (Fig. 5a,b) show 1 to 2-degree
changes from the baseline GCM scenario (1975–1995) for most
of the country by 2030 and increases of over 4 degrees by the end
of the century. A map temperature change from baseline levels
(1975–1990) predicted by the Hadley Centre GCM (Fig. 6a)
shows a very similar level of warming across the country by the
end of the century.
Percentage change in total summertime rainfall based on the
Canadian Climate Centre scenarios are in Fig. 5c,d. This value,
expressed as a percentage, represents the difference between
total future summertime rainfall (May through August) and total
rainfall from the baseline scenario all divided by the mean total
summertime rainfall amount from the baseline GCM scenario
(100 ×[RAINfuture RAINcurrent]/RAINcur rent); thus, positive
values indicate an increase in rainfall in the future in a region.
This ratio is expressed as a percentage change from the base-
line value in Fig. 5c,dand shows a complex pattern of areas
with increased rainfall mixed with areas of decrease in the 2030
scenario. By the end of the 21st century, the Canadian Climate
Centre’s 2090 scenario shows a general increase in rainfall across
the country. Examining seasonal rainfall change in the 2090
scenario of the HAD model compared with its baseline sce-
nario shows a general increase in summertime rainfall across
the country in the future (Fig. 6b) though the pattern of these
262 Int. J. Wildland Fire B. M. Wotton et al.
1 10
Observed number of fires
Predicted number of fires
100 1000
1 10
Observed number of fires
Predicted number of fires
100 1000
Fig. 4. Predicted number of fires v. observed for (a) human- (r2=0.87, P<0.0001) and (b) lightning-
caused (r2=0.76, P<0.0001) fire occurrence models developed here. Each point represents number
of fires per year in an ecoregion.
Fire occurrence and climate change Int. J. Wildland Fire 263
Percentage change
Percentage change
T (°C)
T (°C)
Fig. 5. Maps of changing summer temperature and rainfall projected from the Canadian Climate Centre general circulation model (GCM). Changes in
temperature for (a) 2030 and (b) 2090 represent difference between seasonal average in the future scenario and the baseline scenario in degrees Celsius.
Rainfall change maps for (c) 2030 and (d) 2090 represent the difference between total future summertime rainfall and total rainfall from the baseline scenario
all divided by the mean total summertime rainfall amount from the baseline GCM scenario (100 ×[RAINfuture RAINcurrent]/RAINcurrent ).
increases is spatially different that that observed in the CCC
model (Fig. 5c,d).
The change in 90th percentile for the FFMC was quite small.
The mean value for all the ecoregions from the baseline GCM
scenario was 86 and values increased in the future GCM scenar-
ios as a whole but overall only by 1.6 points (range was 0.9
to 2.6) by the end of the century in the HAD model and by 0.7
point (range was 0.7 to 2.3) in the CCC model. When converted
to actual litter moisture content (using the standard FWI System
relationship), these changes correspond to changes in actual fuel
moisture content of roughly 1.6 and 0.7% respectively.
Changes in 90th percentile DMC values were more spatially
variable than 90th percentile values of FFMC across the country
and greater in terms of their absolute level of change. Across
the ecoregions studied between the baseline and 2090 scenario
from the HAD model, the 90th percentile DMC increased from
28 to 39, while for the corresponding time periods in the CCC
model the 90th percentile DMC values increased from 38 to 49.
These differences amount to an absolute drying of the forest floor
by 30% gravimetric moisture content. The percentage relative
change in 90th percentile DMC level is in Fig. 7 for each of
the future scenarios, and reveals the spatial variability in these
changes across the country.
Tables 2 and 3 summarize the current and future annual
human- and lightning-caused fire occurrence rates for the
forested region of each province and territory studied. Table 2
shows that the baseline GCM scenarios tended to under-predict
human-caused fire occurrence rates across the country. This
is most likely because GCM baseline scenarios tended to be
somewhat wetter that than the current climate, lowering fuel
moisture levels and leading to lower than expected number of
fires. Results from the baseline GCM scenarios and the light-
ning models (Table 3) showed similar under-prediction from the
Hadley Centre model while numbers of fires from the Canadian
Climate Centre were similar to those observed. This agreement
would suggest DMC levels in the CCC baseline scenario were
264 Int. J. Wildland Fire B. M. Wotton et al.
Percentage change
T (°C)
Fig. 6. Maps of changing temperature and rainfall projected from the Hadley Centre general circulation model (GCM) for
the 2090 time slice. Changes in (a) temperature are in degrees Celsius and represent difference between seasonal average in
the future scenario and the baseline scenario. (b) The rainfall change map for 2090 represents the difference between total
future summertime rainfall and total rainfall from the baseline scenario all divided by the mean total summertime rainfall
amount from the baseline GCM scenario (100 ×[RAINfuture RAINcurrent]/RAINcurrent).
Fire occurrence and climate change Int. J. Wildland Fire 265
change in DMC
Fig. 7. Percentage change in 90th percentile level of DMC value for (a) the CCC 2030 scenario, (b) the CCC 2090 scenario and (c) the HAD 2090 scenario.
comparable to actual values for this period. This would imply
that rainfall differences between the current observed weather
and CCC baseline were comprised mainly of more frequent small
events, under the 1.5 mm rainfall threshold below which DMC is
not influenced by rainfall. Overall the under-prediction seemed
strongest in the Hadley Centre model; however, as with previous
studies of climate change and fire danger, where wet scenarios
reduced absolute danger rating levels (e.g. Flannigan et al. 1998,
2000), we considered results from future scenarios always in
relation to those initial predictions from the baseline scenarios
(percentage change from baseline).
Overall, increase in fire occurrence from the baseline CCC
GCM scenario is shown for the 2030 and 2090 time periods for
human and lightning fire cause groups in Fig. 8. Fig. 9 shows
the equivalent maps for the HAD GCM scenarios. Increases in
overall fire occurrence for each of the future scenarios from the
Canadian Climate Centre and Hadley Centre GCMs are summa-
rized by each province in Table 4. Overall, the Hadley Centre
model shows larger increases in forest fire occurrence across
the country, predicting an overall increase in fire activity by the
end of the 21st century of just under 150%. The corresponding
overall increase in fire occurrence from the Canadian Climate
Centre based projections was 74%; the increase for the 2030
time slice was 25%. In the study by Flannigan et al. (2005)
using the same GCM scenarios, projected area burned for the
country increased by about 75% by the end of the century using
the CCC scenario and 150% for the same time period from the
HAD scenario. These values agree quite closely with changes
in fire occurrence levels found here. Increases in area burned in
the Flannigan et al. (2005) study were greatest at northern lati-
tudes across the country in the CCC scenario, while Fig. 8 shows
increases in both human and lightning fire activity are greatest
in central Canada. The results in Flannigan et al. (2005) for the
Hadley model showed expect increases across the entire coun-
try, whereas in terms of fire occurrence, we see strong increases
mainly through the southern and central boreal forest sections
of the country.
Overall output from both GCMs shows increased dryness in fuel
moisture leads to increased fire activity across the country. This
266 Int. J. Wildland Fire B. M. Wotton et al.
2030 Lightning
2090 Lightning
2030 Human
2090 Human
Fig. 8. Relative change (percentage increase) in fire occurrence between future and baseline scenarios for the Canadian Climate Centre general circulation
model (GCM). Relative change is given as the percentage increase in number of fires predicted by the GCM (future scenario minus baseline scenario) divided
by the total number of fires in the baseline scenario.
increase in fire occurrence projected from the output of the two
GCMs is driven most strongly by increases in lightning fire activ-
ity. Across the country, the Canadian Climate Centre results show
human-caused fire increasing by just 16% in the 2030 period
and 36% in the 2090, with corresponding changing in lightning
fire occurrence rates at 31 and 98% respectively. The overall
increase in fire occurrence from the Hadley Centre model is 88%
for human-caused fire and almost 200% for lightning-caused
fires. Given the under-prediction in baseline fire occurrence
by the fire weather generated from the Hadley Centre model,
these results should be interpreted with some degree of caution.
Wotton et al. (2003) developed human-caused fire occurrence
models for ecoregions within the province of Ontario.Their pro-
jections of increases in fire occurrence, using both the CCC and
HAD model (50 and 77%), agree with increases in projected
for Ontario in this current work. In a further study in Ontario,
Wotton et al. (2005) used a very conservative approach to mod-
elling lightning fire occurrence across the province and projected
an 80% increase in lightning fires by the end of the 21st century:
projected values for Ontario from this study (Table 3) are three
times this increase. Krawchuk et al. (2009) used RCM output
(from the Canadian RCM, which is initialized with output from
the CCC GCM: Laprise et al. 2003) to estimate potential change
in lightning fire occurrence in a study area in north-eastern
Alberta. They projected an increase in lightning fire occurrence
of 80%. The average increase in lightning fire occurrence for
Alberta in this study is 30% (CCC model) and 110% (HAD
model), though the ecoregion specific maps reveal that in the
area of the Krawchuk et al. (2009) study the agreement between
results of that study and this current work are probably closer.
Examining potential regional changes, the Canadian Cli-
mate Centre model and Hadley Centre model both show strong
increases in lightning fire activity through Ontario and into
southern Manitoba. Strong increases in lightning fire activity in
Canada’s Northwest Territories (NWT) appear only in the results
of the Canadian Climate Centre model, reflecting the significant
Fire occurrence and climate change Int. J. Wildland Fire 267
Percentage change
from baseline
Lightning 2090
Human 2090
Fig. 9. Relative change (percentage increase) in fire occurrence between future and baseline scenarios using the
Hadley Centre general circulation model (GCM). Relative change is given as the percentage increase in number
of fires as predicted solely by the GCM; that is, future scenario (2080 to 2099) minus baseline scenario (1975 to
1990) divided by the total number of fires in the baseline scenario (1975–1990).
268 Int. J. Wildland Fire B. M. Wotton et al.
Table 4. Projected total change (percentage above current levels) in forest fire occurrence
expected in each province or territory using the projections of the Canadian Climate Centre
and Hadley Centre models
Province or territory Projected increase (%) in total annual fire occurrence rate
Canadian Climate Centre Hadley Centre
2020–2040 2080–2100 2080–2099
British Columbia 9 21 190
Alberta 11 26 87
Saskatchewan 8 22 82
Manitoba 34 85 87
Ontario 46 180 150
Quebec 8 33 140
Yukon 5 24 26
Northwest Territories 46 43 30
Total 25 74 144
increase in rainfall the Hadley model generates in the north of
Canada. This increase in lightning fire activity shown by the
Canadian Climate Centre model agrees with other studies of
changing fire danger, which show the near-arctic in Canada as
being an area particularly vulnerable to climate change, and a
likely location to see initial effects of the changing climate (Soja
et al. 2006).
There are clearly some significant differences in the projec-
tions of the Canadian Climate Centre and the Hadley Centre
GCMs at the regional level that have contributed to the overall
difference in the projections of fire occurrence. Fig. 6bshows
that the regional pattern of rainfall increase across Canada pro-
jected by the HAD model differs from that projected by the CCC
model (Fig. 5d). It would reasonable to assume that these dif-
ferences in rainfall contribute to a large part of the differences
in the patterns of overall fire occurrence change between the
two CMs, given the relative consistency of the spatial patterns
in future temperature (Figs 5b,6a). Model intercomparisons of
major GCMs (including both the Hadley and Canadian Climate
Centre models) used as part of IPCC assessment reports (Meehl
et al. 2007; Randall et al. 2007) have shown that the models
typically agree on temperature at large time and space scales
but there is less consensus on precipitation. This divergence
between GCM outputs is common in GCM intercomparisons and
would be most evident when comparing regional differences. A
GCM scenario of future climate is just a plausible representation
of future climate given the assumptions and parameterizations
within the physical system being modelled. In this regard, it is
useful to think of the two GCM outputs as providing two possi-
ble representations of future conditions, both wetter than current
for the most part but with different spatial distributions for that
The Hadley model tends to predict larger increases in light-
ning fire activity through the southern parts of the Prairie
Provinces (Alberta, Saskatchewan and Manitoba) and Quebec
than are seen in the corresponding scenarios from the Cana-
dian Climate Centre. The difference between the Hadley Centre
and Canadian Climate Centre predictions is most extreme in
the province of British Columbia; however, GCM outputs for
this area (and the subsequent fire occurrence projections) must
be interpreted with considerable caution as the large grid cell
structure used in the GCMs does not resolve the true elevation
of the mountains. Thus, GCM projections represent fire weather
at average elevations across a large cell (400 ×400 km),
whereas fire weather station outputs typically characterize the
fire weather in forested valley bottom areas. Flannigan et al.
(2005) found the same strong differences (between these two
GCMs) in their future projections of area burned for British
Columbia and parts of Alberta.
It is particularly important in climate change impacts studies to
review assumptions and uncontrolled sources of variability that
may have some influence on projections made. The projections
presented here are based on two well established and internation-
ally used GCM models. Differences in fire climate projected by
these two models lead to different projected levels of increase
in fire activity in some regions. It is likely that differences in
rainfall patterns drive a great deal of the regional differences
between the two models. However, a detailed investigation of
these patterns was beyond the scope of this study. The future fire
occurrence scenarios derived from these two GCMs should be
regarded as two possible realizations of possible fire occurrence
in a future with climate change: it is important to remember that
they do not represent specific predictions of the future.
Neither of our fire occurrence models accounts for detailed
forest type information. The use of ecoregion and the inclusion
of ecoregion interactions with the key moisture content variables
can account for coarse scale vegetation differences; however, for-
est type change (which one can readily assume would accompany
climate change) are not accounted for in these model projections.
For the boreal forests of Canada, this lack of change in forest
composition seems reasonable for projections of activity over the
next several decades, as only a small fraction of the forest is dis-
turbed each year; however, examining the impact of forest change
should be considered a crucial aspect of future studies of long-
term impacts of climate change on Canadian forest fire regimes.
The lightning-caused fire occurrence models presented here
were developed without using lightning as a predictor or in the
estimation of expected number of ignitions (as in Wotton and
Fire occurrence and climate change Int. J. Wildland Fire 269
Martell 2005). This leads to an increased level of variability in
the predictions from these models (compared with, for example,
the fire management operations focussed models of Wotton and
Martell 2005). Thus, these projections do not explicitly account
for changes in lightning activity that is projected to increase
under climate change (Price and Rind 1994; Arif 2006). How-
ever, rainfall intensity was used as a surrogate for lightning
presence and thus, an increase in rainfall rate may act in this
model to indicate an increase in lightning activity accompany-
ing climate change. We have also capped DMC values in future
projections with historical ecoregion maximums to avoid unre-
alistically large projections due to the open-ended nature of the
DMC. This would tend to make projections slightly conserva-
tive, though for the most part across the country, DMC values
were not unrealistically high in future scenarios. Furthermore,
we weigh this potential introduction of conservative estimates
against the danger of extrapolating beyond the environmental
conditions under which models were developed and feel this
assumption was reasonable.
In terms of human-caused fires, rates of occurrence and their
spatial distribution can be influenced by demographics, land use,
etc. (e.g. Vega-Garcia et al. 1995). We have chosen to use only
the most recent 20–25 years of fire data available to limit major
changes in patterns of human activity in the forested areas stud-
ied. The basic implicit assumption in our models is that the basic
social factors governing the presence of human ignition sources
on the landscape do not change with time. As with the assumption
of static forest type, in the short-term (decades) this assumption
seems reasonable, however in the long-term, it is reasonable to
expect significant changes in the social elements influencing
human-caused fire ignition patterns.
The grid cell resolution of the GCMs is quite large, in the
order of 400 km per grid cell side; thus, the mountains on the
western coast of Canada are not well resolved. Projections for
the province of British Columbia and theYukon should be inter-
preted with extreme caution. In such regions where fine scale
weather patterns are strongly influenced by local topography,
detailed regional impacts studies (e.g. Nitschke and Innes 2008)
are needed, and the use of output finer scale models, such as the
Canadian Regional Climate Model (Laprise et al. 2003) should
be considered.
We used two GCMs to develop projections of future fire occur-
rence levels across Canada. While fire activity is projected to
increase across all forested regions studied, the relative increase
in number of fires varies regionally. Overall across Canada, our
results from the Canadian Climate Centre GCM scenarios sug-
gest an increase in overall fire occurrence of 25% by 2030 and
75% by the end of the century. Results projected from fire cli-
mate scenarios derived from the Hadley Centre GCM suggest
fire occurrence will increase by 150% by the end of the cen-
tury. These general increases in fire occurrence across Canada
agreed with national predictions of increases in area burned
under climate change (Flannigan et al. 2005).
Fires are a significant and natural element of the boreal forests
of Canada. Understanding the impact of climate change on for-
est fire activity is important for understanding long-term change
in forests, as well as the size of, and potential emissions from,
terrestrial carbon stocks (Flannigan et al. 2008). Studies examin-
ing all aspects of potential change to the fire regime are important
to develop a full understanding of what the forest landscape may
look like in the future, both with and without human manage-
ment of landscapes. The numbers of fires occurring in a region
not only influences potential area burned on the landscape but is
extremely important to forest fire management as it defines the
load on suppression resources a fire management agency will
face. Throughout the managed forests of Canada, fires are sup-
pressed and in most of cases, kept to a very small size; it is those
fires that escape initial attack that lead to area burned with fires
>200 ha in Canada making up only 3% of fires but accounting
for over 97% of the area burned (Stocks et al. 2002). The g reater
the fire load in an intensively protected fire management zone,
the greater the need for resources. When resource capacities are
exceeded, fires can escape and grow large and it is important to
understand these fires, as they are critical to understanding to
total number of large fires on a managed landscape, and hence
total area burned. Thus, it is important to continue this work with
an exploration of potential future rates of escape fire occurrence.
Studies of initial attack system failure in the province of Ontario
(Wotton et al. 2005) have shown that under climate change,
increased fire occurrence rates lead to even greater increases
in escape fire rates. Such findings will be critical for forest fire
management agencies trying to plan fire management strategy
under climate change.
Datasets used in this analysis have been obtained from various provin-
cial forest fire management agencies throughout Canada over several years
(for numerous projects). The authors thank each of these organizations for
their contribution and collaboration. The late Bernie Todd (Canadian Forest
Service) was instrumental in assembling large portions of these provincial
forest fire datasets and creating a common set of attributes that could be
comparable. It was Bernie Todd who originally held discussions when this
national analysis began. Fire weather streams based on Environment Canada
weather station data come from previous work and were assembled with
the grateful assistance of Walter Skinner of the Meteorological Service of
Canada (Environment Canada). We also acknowledge the Canadian govern-
ment’s Program for Energy Research and Development for support of this
Anderson KR (2000) Predicting convective rainfall amounts from lightning
flash density. In ‘Proceedings 20th Conference on Severe Local Storms’.
(American Meteorological Society: Boston, MA)
Anderson KR (2002) A model to predict lightning-caused fire occur-
rences. International Journal ofWildland Fire 11, 163–172. doi:10.1071/
Anisimov OA, Vaughan DG, CallaghanT, Furgal C, Marchant H, Prowse T,
Hjalmar Viljalmasson H, Walsh J (2007) Polar regions (Arctic and
Antarctic). In ‘Climate change 2007: impacts, adaptation and vulnerabil-
ity. Contribution of Working Group II to the Fourth Assessment Report
of the Intergovernmental Panel on Climate Change’. (Eds ML Parry, OF
Canziani, JP Palutikof, PJ van der Linden, CE Hanson) pp. 653–685.
(Cambridge University Press: Cambridge, UK)
Arif AA (2006) The cloud-to-ground lightning parameterization develop-
ment with the Canadian Regional Climate Model. PhD thesis, Université
du Québec à Montréal, Canada.
Balshi MS, McGuire AD, Duffy P, Flannigan M, Walsh J, Melillo J
(2008) Assessing the response of area burned to changing climate in
270 Int. J. Wildland Fire B. M. Wotton et al.
western boreal North America using a Multivariate Adaptive Regres-
sion Splines (MARS) approach. Global Change Biology 15, 578–600.
Bergeron Y, Flannigan MD (1995) Predicting the effects of climate change
on fire frequency in the southeastern Canadian boreal forest. Water, Air,
and Soil Pollution 82, 437–444. doi:10.1007/BF01182853
Beverly JL,Wotton BM (2007) Modelling the probability of sustained flam-
ing: predictive value of fire weather index components compared with
observations of site weather and fuel moisture conditions. International
Journal of Wildland Fire 16(2), 161–173. doi:10.1071/WF06072
Countryman CM (1972) The fire environment concept. USDA Forest
Service, Pacific Southwest Range and Experiment Station.
(Berkeley, CA)
Covey C, AchutaRao KM, Cubasch U, Jones P, Lambert SJ, Mann ME,
Phillips TJ, Taylor KE (2003) An overview of results from the Cou-
pled Model Intercomparison Project. Global and Planetary Change 37,
103–133. doi:10.1016/S0921-8181(02)00193-5
Drever CR, BergeronY, Drever MC, Flannigan M, Logan T, Messier C (2008)
Effects of climate on occurrence and size of large fires in a northern
hardwood landscape: historical trends, future predictions, and implica-
tions for climate change in Témiscamingue, Québec. AppliedVegetation
Science 12(3), 261–272. doi:10.1111/J.1654-109X.2009.01035.X
Ecological Stratification Working Group (1996) ‘A national ecological
framework for Canada.’Agriculture and Agric.-Food Canada, Research
Branch, Centre for Land and Biological Resources Research and
Environment Canada, State of Environment Directorate. (Ottawa, ON)
Field CB, Mortsch LD, Brklacich M, Forbes DL, Kovacs P, Patz JA,
Running SW, Scott MJ (2007) North America. In ‘Climate Change 2007:
impacts, adaptation and vulnerability. Contribution of Working Group
II to the Fourth Assessment Report of the Intergovernmental Panel on
Climate Change’. (Eds ML Parry, OF Canziani, JP Palutikof, PJ van
der Linden, CE Hanson) pp. 617–652. (Cambridge University Press:
Cambridge, UK)
Flannigan MD, Van Wagner CE (1991) Climate change and wildfire in
Canada. Canadian Journal of Forest Research 21, 66–72. doi:10.1139/
Flannigan MD, Wotton BM (1989) A study of interpolation methods for
forest fire danger rating in Canada. Canadian Journal of Forest Research
19, 1059–1066. doi:10.1139/X89-161
Flannigan MD, Wotton BM (1991) Lightning ignited forest fires in north-
western Ontario. Canadian Journal of Forest Research 21(3), 277–287.
Flannigan MD, Wotton BM (2001) Climate, weather, and area burned.
In ‘Forest fires: behavior and ecological effects’. (Eds EA Johnson,
K Miyanishi) pp. 351–373. (Academic Press: San Diego, CA)
Flannigan MD, BergeronY, Engelmark O, Wotton BM (1998) Future wild-
fire in circumboreal forests in relation to global warming. Journal of
Vegetation Science 9, 469–476. doi:10.2307/3237261
Flannigan MD, Stocks BJ, Wotton BM (2000) Forest fires and cli-
mate change. The Science of the Total Environment 262(3), 221–229.
Flannigan MD, LoganKA, Amiro BD, Skinner WR, Stocks BJ (2005) Future
area burned in Canada. Climatic Change 72, 1–16. doi:10.1007/S10584-
Flannigan MD, Stocks BJ, Turetsky M, Wotton BM (2008) Impacts of
climate change on fire activity and fire management in the circum-
boreal forest. Global Change Biology 15(3), 549–560. doi:10.1111/
Flato GM, Boer GJ, Lee WG, McFarlane NA, Ramsden D, Reader MC,
Weaver AJ (2000) The Canadian Centre for Climate Modelling andAna-
lysis Global Coupled Model and its climate. Climate Dynamics 16(6),
451–467. doi:10.1007/S003820050339
Forestry Canada Fire Danger Group (1992). Development and structure of
the Canadian forest fire behavior prediction system. Forestry Canada,
Science and Sustainable Development Directorate, Special Report
ST-X-3. (Ottawa, ON)
Frandsen WH (1987)The inf luence of moisture and mineral soil on the com-
bustion limits of smouldering duff. Canadian Journal of Forest Research
17, 1540–1544. doi:10.1139/X87-236
Gosz JR, Moore DI, Shore GA, Grover HD, Rison W, Rison C (1995)
Lightning estimates of precipitation location and quantity on the
Sevilleta LTER, New-Mexico. Ecological Applications 5, 1141–1150.
Hulme M, Mitchell J, Ingram W, Lowe J, Johns T, New M, Viner
D (1999) Climate change scenarios for global impacts studies.
Global Environmental Change 9, S3–S19. doi:10.1016/S0959-3780(99)
IPCC (1995) ‘Climate Change 1995: the science of climate change, a
contribution of Working Group I to the Second Assessment of the Inter-
governmental Panel on Climate Change.’ (Eds JT Houghton, LG Meira
Filho, BA Callender, N Harris, A Kattenberg, K Maskell) (Cambridge
University Press: UK)
IPCC (2001) ‘Climate change 2001: synthesis report; a contribution of
Working Groups I, II, and III to the third assessment report of the Inter-
governmental Panel on Climate Change.’ (Eds RT Watson and the Core
Writing Team) (Cambridge University Press: Cambridge: UK)
IPCC (2007) ‘Climate Change 2007: synthesis report. Contribution of
Working Groups I, II and III to the Fourth Assessment Report of the Inter-
governmental Panel on Climate Change.’(Eds RK Pachauri, A Reisinger,
and the Core Writing Team) (IPCC: Geneva, Switzerland)
Johnson EA (1992) ‘Fire and vegetation dynamics.’(Cambridge University
Press: Cambridge, UK)
Kourtz PH, Todd JB (1992) Predicting the daily occurrence of lightning-
caused forest fires. Petawawa National Forestry Institute, Information
Report PI-X-112. (Chalk River, ON)
Krawchuk MA, Cumming SG, Flannigan MD, Wein RW (2006) Biotic and
abiotic regulation of lightning fire initiation in the mixedwood boreal
forest. Ecology 87, 458–468. doi:10.1890/05-1021
Krawchuk MA, Cumming SG, Flannigan MD (2009) Predicted changes
in fire weather suggest increases in lightning fire initiation and future
area burned in the mixedwood boreal forest. Climatic Change 92,
Laprise R, Caya D, Frigon A, Paquin D (2003) Current and perturbed cli-
mate as simulated by the second-generation Canadian Regional Climate
Model (CRCM-II) over northwestern NorthAmerica. Climate Dynamics
21, 405–421. doi:10.1007/S00382-003-0342-4
Lawson BD, Armitage OB (1997) Ignition probability equations for some
Canadian fuel types. Ember Research Services Ltd., Final Report
submitted to the Canadian Committee on Forest Fire Management.
(Victoria, BC)
Lawson BD, Dalymple GN, Hawkes BC (1997) Predicting forest floor
moisture from duff moisture code values. Natural Resources Canada,
Canadian Forest Service, Pacific Forestry Centre, Forest Research
Applications Technology Transfer Note #6. (Victoria, BC)
Lempriere TC, Bernier PY, Carroll AL, Flannigan MD, Gilsenan RP,
McKenney DW, Hogg EH, Pedlar JH, Blain D (2008) The impor-
tance of forest sector adaptation to climate change. Canadian Forest
Service, Northern Forestry Centre, Information Report NOR-X-416E.
(Edmonton, AB)
Logan KA, Flannigan MD, Wotton BM, Stocks BJ (2004) Development
of daily weather and fire danger scenarios using two General Circu-
lation Models. In ‘Proceedings of the 22nd tall timbers fire ecology
conference: fire in temperate, boreal, and montane ecosystems’. (Eds
RT Engstrom RT, WJ de Groot) pp. 185–190. (Tall Timbers Research
Station: Tallahassee, FL)
Mandallaz D,Ye R (1997) Prediction of forest fires with a Poisson models.
Canadian Journal of Forest Research 27(10), 1685–1694. doi:10.1139/
Fire occurrence and climate change Int. J. Wildland Fire 271
Martell DL, Sun H (2008) The impact of forest fire suppression, vegetation
and weather on burned area in Ontario. Canadian Journal of Forest
Research 38(6), 1547–1563. doi:10.1139/X07-210
Martell DL, Otukol S, Stocks BJ (1987) A logistic model for predicting
daily people-caused forest fire occurrence in Ontario. Canadian Journal
of Forest Research 17(5), 394–401. doi:10.1139/X87-068
Martell DL, Bevilacqua E, Stocks BJ (1989) Modelling seasonal variation in
daily people-caused forest fire occurrence. Canadian Journal of Forest
Research 19(12), 1555–1563. doi:10.1139/X89-237
McAlpine RS, Hirsch KG (1999) An overview of LEOPARDS: the level of
protection analysis system. Forestry Chronicle 75, 615–621.
McCullagh P, Nelder J (1989) ‘Generalized linear models.’ (Chapman and
Hall: London, UK)
Meehl GA, Covey C, Delworth T, Latif M, McAvaney B, Mitchell JFB,
Stouffer RJ, Taylor KE (2007) The WCRP CMIP3 multimodel dataset:
a new era in climate change research. Bulletin of the American Meteo-
rological Society 88(9), 1383–1394. doi:10.1175/BAMS-88-9-1383
Nitschke CR, Innes JL (2007) Climate change and fire potential in
south-central British Columbia, Canada. Global Change Biology 14(4),
841–855. doi:10.1111/J.1365-2486.2007.01517.X
Otway SG, Bork EW, Anderson KR, Alexander ME (2007) Relating changes
in duff moisture to the Canadian Forest Fire Weather Index System
in Populus tremuloides stands in Elk Island National Park. Canadian
Journal of Forest Research 37, 1987–1998. doi:10.1139/X07-055
Pew KL, Larsen CPS (2001) GIS analysis of spatial and temporal patterns
of human-caused wildfires in the temperate rain forests of Vancouver
Island, Canada. Forest Ecology and Management 140, 1–18.
doi:10.1016/ S0378-1127(00)00271-1
Poulin-Costello M (1993) People-caused forest fire prediction using Poisson
and logistic regression. MSc thesis, University ofVictoria, Victoria, BC,
Price C, Rind D (1994) The impact of a 2×CO2climate on lightning-
caused fires. Journal of Climate 7, 1484–1494. doi:10.1175/1520-
Randall DA, Wood RA, Bony S, Colman R, Fichefet T, Fyfe J, Kattsov V,
Pitman A, Shukla J, Srinivasan J, Stouffer RJ, Sumi A, Taylor KE (2007)
Climate models and their evaluation. In ‘Climate change 2007: the
physical science basis. Contribution of Working Group I to the Fourth
Assessment Report of the Intergovernmental Panel on Climate Change’.
(Eds S Solomon, D Qin, M Manning, Z Chen, M Marquis, KB Averyt,
M Tignor, HL Miller) (Cambridge University Press: Cambridge, UK)
Raupach MR, Marland G, Ciais P, Le Quéré C, Canadell JG, Klepper G, Field
CB (2007) Global and regional drivers of accelerating CO2emissions.
Proceedings of the NationalAcademy of Sciences of the United States of
America 104, 10 288–10 293. doi:10.1073/PNAS.0700609104
Roulet N, Moore T, Bubier J, Lafleur P (1992) Northern fens – methane flux
and climatic change. Tellus B 44, 100–105. doi:10.1034/J.1600-0889.
Sheridan SC, Griffiths JF, Orville RE (1997) Warm season cloud-to-ground
lightning-precipitation relationships in the south-central United States.
Weather and Forecasting 12, 449–458. doi:10.1175/1520-0434(1997)
Soja AJ, Tchebakova NM, French NHF, Flannigan MD, Shugart HH,
Stocks BJ, SukhininAI, Parfenova EI, Chapin FS, III, Stackhouse PW, Jr
(2006) Climate-induced boreal forest change: predictions versus current
observations. Global and Planetary Change 56, 274–296. doi:10.1016/
Stocks BJ (1993) Global warming and forest fires in Canada. Forestry
Chronicle 69, 290–293.
Stocks BJ, Lawson BD, Alexander ME, Van Wagner CE, McAlpine RS,
Lynham TJ, Dube DE (1989) The Canadian forest fire danger rating
system: an overview. Forestry Chronicle 65, 450–457.
Stocks BJ, Fosberg MA, Lynham TJ, Mearns L, Wotton BM, Yang Q, Jin
J-Z, Lawrence K, et al. (1998) Climate change and forest fire potential
in Russian and Canadian boreal forests. Climatic Change 38(1), 1–13.
Stocks BJ, Mason JA, Todd JB, Bosch EM, Wotton BM, Amiro BD,
Flannigan MD, Hirsch KG, Logan KA, Martell DL, Skinner WR
(2002) Large forest fires in Canada, 1959–1997. Journal of Geophysical
Research 107, 8149. doi:10.1029/2001JD000484
SwetnamTW (1993) Fire history and climate change in giant sequoia g roves.
Science 262, 885–889. doi:10.1126/SCIENCE.262.5135.885
Tapia A, Smith JA, Dixon M (1998) Estimation of convective rainfall from
lightning observations. Journal of Applied Meteorology 37, 1497–1509.
Ty m s t r a C , F l a n n i g a n M , A r m i t a g e B, L o g a n K ( 2 0 0 7 ) I m p a c t o f c l i m a t e
change on area burned in Alberta’s boreal forest. International Journal
of Wildland Fire 16, 153–160. doi:10.1071/WF06084
Van Wagner CE (1972) Duff consumption by fire in eastern pine stands.
Canadian Journal of Forest Research 2, 34–39. doi:10.1139/X72-006
Van Wagner CE (1987)The development and structure of the Canadian Forest
Fire Weather Index System. Canadian Forest Service, Petawawa National
Forestry Institute, ForestryTechnical Report FTR-35. (Chalk River, ON)
Vega -G ar c ia C , Woo da r d PM , Titu s SJ, A dam ow ic z WL , Le e B S (1 99 5)
A logit model for predicting the daily occurrence of human caused
forest fires. International Journal of Wildland Fire 5(2), 101–112.
Weber MG, Stocks BJ (1998) Forest fires and the sustainability in the boreal
forests of Canada. Ambio 27, 545–550.
Wilmore B (2001) Duff moisture dynamics in black spruce feather moss
stands and their relation to the Canadian Forest Fire Danger Rating
System. MSc thesis, University of Alaska, Fairbanks.
Wotton BM (2009) Inter preting and using outputs from the Canadian forest
fire danger rating system in research applications. Environmental and
Ecological Statistics 16(2), 107–131. doi:10.1007/S10651-007-0084-2
Wotton BM, Beverly JL (2007) Stand specific litter moisture content cali-
brations for the Canadian fine fuel moisture code. International Journal
of Wildland Fire 16(4), 463–472. doi:10.1071/WF06087
Wotton BM, Flannigan MD (1993) Length of the fire season in a changing
climate. Forestry Chronicle 69, 187–192.
Wotton BM, Martell DM (2005) A lightning fire occurrence model for
Ontario. Canadian Journal of Forest Research 35(6), 1389–1401.
Wotton BM, Stocks BJ (2006) Fire management in Canada: vulnerability
and risk trends. In ‘Canadian wildland fire strategy: background synthe-
sis, analysis, and perspectives’. (Eds KG Hirsch, P Fuglem) Canadian
Council of Forest Ministers, Natural Resources Canada, Canadian Forest
Service, Northern Forestry Centre, pp. 49–55. (Edmonton, AB)
Wotton BM, Stocks BJ, Flannigan MD, Laprise R, Blanchet JP (1998) Esti-
mating current and future fire climates in the boreal forest of Canada
using a regional climate model. In ‘Proceedings of the 3rd international
conference on forest fire research and 14th conference on fire and for-
est meteorology’, Coimbra, Portugal. (Ed. DX Viegas) pp. 1207–1221.
(Elsevier: Amsterdam, the Netherlands)
Wotton BM, Martell DM, Logan KA (2003) Climate change and people-
caused forest fire occurrence in Ontario. Climatic Change 60, 275–295.
Wotton BM, Logan KA, McAlpine RS (2005) Climate change and the
future fire environment in Ontario: fire occurrence and fire management
impacts. Ontario Ministry of Natural Resources, Ontario Forest Research
Institute, Climate Research Report CCRR-01. (Sault Ste. Marie, ON)
Manuscript received 8 January 2009, accepted 12 October 2009
... Early start, delayed end, and increased length of forest fire season are also believed to be happening [1]. Occurrences of the number of fires and changes in the fire season in a region are considered as critical factors for fire management agencies to determine the necessary fire suppression resources [5,6]. Such resources are usually developed based on the existing knowledge, experience, and understanding of fire occurrence trends and seasonality. ...
... Therefore, when even a small number of forest fires exceed the historical number or the fire season extends in a region, it might transcend the existing resource capacities [5,7]. It could lead more escaped fires, and potentially cause a larger burned area [6]. In fact, climate change (mostly warming; [8]) and more human activities in the forest [9,10] are considered responsible for the changes in fire weather that lead to the increasing trends of forest fire occurrences and longer season. ...
... Fire 2023, 6, 96 ...
Full-text available
We analyzed the distribution and number of forest fire occurrences, burned areas, and seasonality, and their trends of human- and lightning-caused small (<200 ha) and large (≥200 ha) fires from 1959 to 2021 in the forested 14 subregions of Alberta, based on the Canadian National Fire Database. We applied a non-parametric statistical test, i.e., Mann–Kendall and Sen’s slope estimator, for the patterns and magnitudes of the trends. Our results revealed that all subregions experienced significantly increasing trends of fire occurrences, either monthly or yearly, except the Alpine subregion. In the burned area case, nine ecoregions demonstrated significantly decreasing monthly trends for small fires caused by humans, except for an increasing trend in the Lower Boreal Highlands subregion in May. For seasonality, we found one to two days for both early start and delayed end of fire season, and eventually two to four days longer fire seasons in five ecoregions. This study provides an updated understanding of the fire regimes in Alberta. It would be helpful for fire management agencies to make strategic plans by focusing on high-priority regions to save lives and properties.
... Wildfire is a major disturbance in Canada's boreal forest, contributing to an average 2 M ha of stand renewal annually (Stocks et al. 2002). Fire activity is anticipated to increase with climate change across boreal forests of Canada (Balshi et al. 2009;Wotton et al. 2010;de Groot et al. 2013;Boulanger et al. 2014;Wang et al. 2017) and may cancel out potential increase in forest stock (Gauthier et al. 2015b;Rapanoela et al. 2015;Beaudoin et al. 2017;Chaste et al. 2019). In the context of sustainable boreal forest management, it seems particularly important to assess the combined impact of climate change via growth changes and changing fire activity. ...
... area becomes more extended. Warmer conditions can lead to a longer vegetative season and thus promote growth of trees (Menzel and Fabian 1999;Chmielewski and Rötzer 2001;Messaoud and Chen 2011;Price et al. 2013;D'Orangeville et al. 2016D'Orangeville et al. , 2018Hember et al. 2017;Moreau et al. 2020) but also leads to favorable conditions for fires, which results in increased fire activity (Balshi et al. 2009;Wotton et al. 2010;de Groot et al. 2013;Boulanger et al. 2014Boulanger et al. , 2017Wang et al. 2017). ...
Full-text available
Climate change is having complex impacts on the boreal forest, modulating both tree growth limiting factors and fire regime. However, these aspects are usually projected independently when estimating climate change effect on the boreal forest. Using a combination of 3 different methods, our goal is to assess the combined impact of changes in growth and fire regime due to climate change on the timber supply at the transitions from closed to open boreal coniferous forests in Québec, Canada. In order to identify the areas that are likely to be the most sensitive to climate change, we projected climate-induced impacts on growth and fire activity at three different time periods: 2011-2040 RCP 8.5 for low growth change and minimum fire activity, 2071-2100 RCP 4.5 for moderate growth change and medium fire activity, and 2071-2100 RCP 8.5 for high growth change and maximum fire activity. Our study shows the importance of incorporating fire in strategic forest management planning especially in a context of climate change. Under the most extreme scenarios the negative impact of fire activity on productive area and total volume mostly offsets the positive effects of climate change via improved tree growth.
... Most of the existing work on fire risk assessment and prediction addresses forest land, forest fires and urban-forest boundaries, such as in China [11], Portugal [12], and Canada [13]. They use different methods such as random forests [11], general circulation models [13], and artificial neural networks [12], aiming to support the allocation of fire protection, fire prevention, and plant restoration resources to areas with the highest fire risk. ...
... Most of the existing work on fire risk assessment and prediction addresses forest land, forest fires and urban-forest boundaries, such as in China [11], Portugal [12], and Canada [13]. They use different methods such as random forests [11], general circulation models [13], and artificial neural networks [12], aiming to support the allocation of fire protection, fire prevention, and plant restoration resources to areas with the highest fire risk. The characteristics used for forest fires, such as soil type, topography, humidity, and rainfall, are very different from those typically used in urban fire assessment and prediction (e.g., combustibility of building materials, fire performance, and type of property use). ...
Full-text available
Machine learning methods can establish complex nonlinear relationships between input and response variables for stadium fire risk assessment. However, the output of machine learning models is considered very difficult due to their complex “black box” structure, which hinders their application in stadium fire risk assessment. The SHapley Additive exPlanations (SHAP) method makes a local approximation to the predictions of any regression or classification model so as to be faithful and interpretable, and assigns significant values (SHAP value) to each input variable for a given prediction. In this study, we designed an indicator attribute threshold interval to classify and quantify different fire risk category data, and then used a random forest model combined with SHAP strategy in order to establish a stadium fire risk assessment model. The main objective is to analyze the impact analysis of each risk characteristic on four different risk assessment models, so as to find the complex nonlinear relationship between risk characteristics and stadium fire risk. This helps managers to be able to make appropriate fire safety management and smart decisions before an incident occurs and in a targeted manner to reduce the incidence of fires. The experimental results show that the established interpretable random forest model provides 83% accuracy, 86% precision, and 85% recall for the stadium fire risk test dataset. The study also shows that the low level of data makes it difficult to identify the range of decision boundaries for Critical mode and Hazardous mode.
... Anthropogenic climate change is altering the fire regimes of Western Canada, where forest fires are increasing in frequency, intensity and size (Flannigan et al., 2019(Flannigan et al., , 2013Wotton et al., 2010;deGroot et al., 2013;Kirchmeyer-Young et al., 2016). The risk of large mega-fires in coniferous forests, such as those experienced in Fort McMurray, is increasing with anthropogenic warming due to drier, warmer spring seasons, and increased lightning storms which ignite fires. ...
Full-text available
This thesis explores processes of opportunism and socio-political change following so-called “natural” disasters through a multi-sited case study of the post-wildfire morel mushroom (Morchella sp.) harvest in Western Canada. Morels are edible fungi which fruit en masse the first spring season following large wildfires in western North America. This work follows harvesters who picked morels after the 2016 Horse River wildfire on Treaty 8 territory (near Fort McMurray, Alberta), and the 2018 Shovel Lake wildfire on Nadleh Whut’en, Stellat’en, and Nak’azdli Whut’en Territories (British Columbia). Thinking with and beyond the concepts of disaster capitalism and disaster colonialism, this thesis extends analyses of post-disaster change from considerations of states and large corporations to smaller-scale actors. As such, I consider the roles of hobbyist local harvesters, precarious pieceworkers in the wild mushroom industry, Indigenous Guardians of the Land, and forest ecologies more broadly. I demonstrate that while disaster capitalism and disaster colonialism are pervasive in the post-disaster landscape, they are not inevitable. Instead, I argue that post-disaster opportunity is emergent, contingent, and includes possibilities for reworking, resistance, and resurgence. In this work, I argue that settler-colonial aims to subsume nature produce the ecological conditions which make the commercial mushroom harvest possible. This industry, in turn, disproportionately benefits settler harvesters over Indigenous Nations and forest ecologies. I also demonstrate that the materiality of wildfire memories affects different groups’ capacities to harvest mushrooms, influence others, and define the ethical standards of the harvest. Finally, I examine how settler claims to post-disaster opportunity on Indigenous lands¬ are connected to broader affective “settler common sense” and “white possessive” claims to adventure, freedom and commerce. Together, these findings demonstrate how the concurrent and often contradicting post-disaster opportunism demonstrated by small-scale actors relate to broader politics about natural disasters, environmental politics, resource extraction, and Indigenous governance within Canada and in other settler-colonial contexts.
... By 2100, the annual area burned in Canada is projected to increase by two to four times (Boulanger et al., 2014;Coogan et al., 2019). Such situations may overwhelm the capacity of fire management agencies (Wotton et al., 2010) and result in a substantial increase in fire management expenditures (Hope et al., 2016). ...
Full-text available
Natural disturbances drive forest dynamics and biodiversity at different spatial and temporal scales. Forests in the boreal biome are shaped by several types of disturbance, including fire, windthrow, and insect outbreaks, that vary in frequency, extent, severity, and specificity. In managed forests, disturbances also affect the amount and quality of available timber. Ecosystem management uses information on disturbance regimes as a guide to finding a balance between ecological, economic, and social viewpoints. In this chapter, we review current knowledge on disturbance regimes in boreal forests and discuss some implications for managing the impact and risk of disturbances in the context of forest ecosystem management and restoration.
... Expected future wildfire challenges in Canada and globally are well documented [4][5][6]. Under the effects of climate change, Canadian wildfires are expected to see increases in frequency [7,8], severity [9,10], and season length [11,12], leading to larger social, economic, and environmental impacts [13][14][15][16]. In addition, there are challenges that will be faced by those who manage wildfire risk; however, there are also a wide variety of options for mitigating and adapting to this risk [17,18]. ...
Full-text available
Wildfires are worsening in Canada and globally, partly due to climate change. The government of Canada is designing and building WildFireSat, the world’s first purpose-built operational satellite system for wildfire monitoring. It will provide new fire intelligence to support decision-making. It takes time for fire management agencies to use new information: to understand it and its implications, change processes, develop training, and modify computer systems. Preparing for the system’s prelaunch will allow agencies to benefit more rapidly from the new information. We present (1) an assessment of the readiness of 12 Canadian fire management agencies to integrate WildFireSat information and (2) guidance for reducing readiness gaps. We used survey and other data to score readiness indicators for three readiness components: understanding, organization, and information technology. We weighted the influence of each indicator score on each component. We modelled scoring and weighting uncertainties and used Monte Carlo simulation to generate distributions of aggregated agency readiness. The results indicated that most agencies have a moderate level of readiness while others have a higher level of readiness. Cluster analysis was used to group agencies by similarity in multiple dimensions. Strategies for increasing readiness are highlighted. This identifies opportunities for agencies and the WildFireSat team to collaborate on enhancing readiness for the forthcoming WildFireSat data products.
Governmental legislation, regulations, and policies are used to prevent and mitigate the negative impact of human-caused wildland fires. In Ontario, Canada, the Modifying Industrial Operations Protocol (MIOP) aims to manage and limit the risk associated with fires ignited because of industrial forestry operations while maintaining flexibility in terms of daily restrictions. The MIOP was enacted in Ontario in 2008, when it replaced the Woods Modifications Guidelines, which had been in effect since 1989. We use quantile regression to quantify how the distribution of incremental growth has changed when contrasting three prevention time periods (MIOP, Woods Guidelines, Pre-Woods) while controlling for several possible confounding variables that drive fire growth. We analyze data of industrial forestry-caused wildland fires ignited on Crown forest land in Ontario from 1976 to 2019. This type of retrospective analysis is important for monitoring the performance of Ontario’s prevention and mitigation efforts and providing insight for the future, especially in a changing environment. Our findings provide evidence of MIOP succeeding at its goal of mitigating the negative impact of ignited industrial forestry fires when compared against previous regulations. Study Implications: Regulations are one avenue for mitigating risk associated with the accidental ignition of fires by forestry operations. The Modifying Industrial Operations Protocol (MIOP) aims to be more flexible than its predecessor, so we investigate whether forestry-caused fires are tending to grow larger or smaller under MIOP compared to previous time periods. Quantile regression allows us to model individual quantiles of the distribution of incremental growth, the difference between a fire’s discovery and final sizes, while controlling for several confounders that influence growth. We find evidence of improvements to the right tail of this distribution, with fires growing less under MIOP.
Full-text available
Abstract Geoprocessing techniques are generally applied in natural disaster risk management due to their ability to integrate and visualize different sets of geographic data. The objective of this study was to evaluate the capacity of classification and regression tree (CART) to assess fire risk. MCD45A1 product of the burnt area, relative to a 16-year period (2000-2015) was used to obtain a fire occurrence map, from center points of the raster, using a kernel density approach. The resulting map was then used as a response variable for CART analysis with fire influence variables used as predictors. A total of 12 predictors were determined from several databases, including environmental, physical, and socioeconomic aspects. Rules generated by the regression process allowed to of define different risk levels, expressed in 35 management units, and used to produce a fire prediction map. Results of the regression process (r = 0.94 and r² = 0.88) demonstrate the capability of the CART algorithm in highlighting hierarchical relationships among predictors, while the model’s easy interpretability provides a solid basis for decision making. This methodology can be expanded in other environmental risk analysis studies and applied to any area of the globe on a regional scale.
Climate change has an impact on forest fire patterns. In the context of global warming, it is important to study the possible effects of climate change on forest fires, carbon emission reductions, carbon sink effects, forest fire management, and sustainable development of forest ecosystems. This study is based on MODIS active fire data from 2001–2020 and the influence of climate, topography, vegetation, and social factors were integrated. Temperature and precipitation information from different scenarios of the BCC-CSM2-MR climate model were used as future climate data. Under climate change scenarios of a sustainable low development path and a high conventional development path, the extreme gradient boosting model predicted the spatial distribution of forest fire occurrence in China in the 2030s (2021–2040), 2050s (2041–2060), 2070s (2061–2080), and 2090s (2081–2100). Probability maps were generated and tested using ROC curves. The results show that: (1) the area under the ROC curve of training data (70%) and validation data (30%) were 0.8465 and 0.8171, respectively, indicating that the model can reasonably predict the occurrence of forest fire in the study area; (2) temperature, elevation, and precipitation were strongly correlated with fire occurrence, while land type, slope, distance from settlements and roads, and slope direction were less strongly correlated; and, (3) based on future climate change scenarios, the probability of forest fire occurrence will tend to shift from the south to the center of the country. Compared with the current climate (2001–2020), the occurrence of forest fires in 2021–2040, 2041–2060, 2061–2080, and 2081–2100 will increase significantly in Henan Province (Luoyang, Nanyang, Sanmenxia), Shaanxi Province (Shangluo, Ankang), Sichuan Province (Mianyang, Guangyuan, Ganzi), Tibet Autonomous Region (Shannan, Linzhi, Changdu), Liaoning Province (Liaoyang, Fushun, Dandong).
Wildland fire is an important natural disturbance in many vegetated areas of the world. However, fire management actions are critical not only to prevent and suppress unwanted fires, but also mitigate and recover from the negative impacts of fire on people and communities. Advancements in wildland fire science can help inform these necessary actions in wildland fire management. How science is created and integrated into these fire management decision-making processes, whether through collaborations with external researchers and/or with scientists within a wildland fire management agency itself, requires a conscious understanding of how the science is useful and goes beyond the simple existence of knowledge. This chapter outlines the goal of integrating fire science and management using a conceptual knowledge exchange (KE) framework, informed from existing work on KE. We provide a review of the KE literature relevant to wildland fire management and develop a KE framework for the fire management context. In this context, we address the potential barriers and facilitators throughout this process followed by a discussion of an active learning approach aimed at developing effective data translation skills amongst students in a data analytics consulting course.KeywordsWildland fireKnowledge exchangeKnowledge transferTechnical transferFire managementFire science
Full-text available
The looming possibility of global wanning raises legitimate concerns for the future ofthe forest resource in Canada. While evidence ofaglobal warming trend is not conclusive at this time, governments would be wise to anticipate, and begin planning for. such an eventuality. The forest fire business islikely tobe affected both early and dramatically by any trend toward wanner and drier conditions in Canada, and fire managers should be aware that the future will likely require new and innovative thinking in forest fire management. This paper summarizes research activities currently underway to assess the impact of global warming on forest fires, and speculates on future fire management problems and strategies.
Open access to an unprecedented, comprehensive coordinated set of global coupled climate model experiments for twentieth and twenty-first century climate and other experiments is changing the way researchers and students analyze and learn about climate. The history of climate change modeling was first characterized in the 1980s by a number of distinct groups developing, running, and analyzing model output from their own models with little opportunity for anyone outside of those groups to have access to the model data. This was partly a consequence of relatively primitive computer networking and data transfer capabilities, along with the daunting task of collecting and storing such large amounts
Typically, 50-70% of the total annual precipitation in New Mexico can be produced by convective thunderstorms during the period June through September. These thunderstorms are accompanied by intense lightning and characteristically produce heavy, localized rainfall resulting in high spatial variation in precipitation inputs. During other months precipitation over the entire Sevilleta ($10^5$ ha) often occurs from broad-scale storm systems and is much less spatially variable on a per-storm basis. Summer precipitation is a primary factor driving plant productivity as well as influencing nutrient cycling, herbivore activity, and detritivore activity. Knowledge of the timing, location, and amounts of precipitation is important in planning or monitoring research activities and spatial modeling of the dynamics in this semiarid region. Technology exists for locating cloud-to-ground lightning strikes that has the potential to locate these intense precipitation events, quantify the volume of water associated with them, and document the spatial and temporal variability of this phenomenon over large areas. Near real-time analysis capability can identify areas receiving precipitation that will experience rapid vegetation growth in this semiarid region. This study developed algorithms relating lightning and precipitation quantity and used lightning location to determine rainfall depth and distribution for areas in New Mexico. There was a significant correlation between rain-gauge measured precipitation and lightning within a 3-km radius of the gauge location, with best predictions occurring from regressions that included lightning strikes and relative humidity. Average precipitation volume per cloud-to-ground lightning strike averaged 36 190 m$^3$ for the 3 km radius circle, resulting in an average rainfall depth of 1.3 mm per lightning strike. Lightning location technology, combined with a Geographic Information System (GIS), defined the spatial and temporal resolution of these intense, summer precipitation patterns and provided a more detailed estimate of total precipitation and precipitation distribution than was provided by the sparse network of precipitation gauges. Combining this information with satellite sensing of vegetation growth (e.g., greenness index) can identify causal mechanisms for temporal and spatial patterns in short-term vegetation processes (e.g., primary production) and long-term vegetation dynamics for this area.
Forest fire danger rating research in Canada was initiated by the federal government in 1925. Five different fire danger rating systems have been developed since that time, each with increasing universal applicability across Canada. The approach has been to build on previous danger rating systems in an evolutionary fashion and to use field experiments and empirical analysis extensively. The current system, the Canadian Forest Fire Danger Rating System (CFFDRS), has been under development by Forestry Canada since 1968. The first major subsystem of the CFFDRS, the Canadian Forest Fire Weather Index (FWI) System, provides numerical ratings of relative fire potential based solely on weather observations, and has been in use throughout Canada since 1970. The second major subsystem, the Canadian Forest Fire Behavior Prediction (FBP) System, accounts for variability in fire behavior among fuel types (predicting rate of spread, fuel consumption, and frontal fire intensity), was issued in interim form in 1984 with final production scheduled for 1990. A third major CFFDRS subsystem, the Canadian Forest Fire Occurrence Prediction (FOP) System, is currently being formulated. This paper briefly outlines the history and philosophy of fire danger rating research in Canada discussing in detail the structure of the current CFFDRS and its application and use by fire management agencies throughout Canada. Key words: fire danger, fire behavior, fire occurrence prediction, fuel moisture, fire danger rating system, fire management.