https://orcid.org/0000-0002-6992-9614
Figures
Abstract
Large wildfires are a dominant type of disturbance in North American forests that have caused significant environmental impacts and economic damages. A potential approach to mitigate wildfire spread risk is to segment the landscape with fuel breaks or similar areas with reduced fuel content. However, fragmenting the landscape may undermine other ecosystem services, such as the provision of wildlife migration routes. We present a mixed-integer programming fuel break planning model that minimizes wildfire hazard while protecting migration routes for an endangered wildlife species population (woodland caribou, Rangifer tarandus caribou). Our model applies the network optimization concept and minimizes the expected burn area in the landscape while protecting a target proportion of the wildlife migration routes. We depict the landscape as a network of patches with fuel and track the impact of fuel breaks across a large set of predicted fire spread scenarios where each scenario is conceptualized as a fire propagation graph. We formulated the fuel break allocation as a scenario-based, mixed-integer optimization problem and used the fuel break widths to track the potential failures of the fuel breaks to contain the fire spread. We also considered a planning problem that minimized the expected area of the largest fires and fuel connectivity in the landscape. For given fire hazard reduction goals, the optimal solutions are those with the least disruptive impacts on the wildlife migration routes. We implemented the approach for the planning of wildfire mitigation in the Red Rock-Prairie Creek area of Alberta (Canada), a fire-prone landscape with the presence of caribou populations. The proposed approach reduced the wildfire spread potential in the area with a trade-off of moderate impacts on known caribou migration routes. Such a methodology is timely to help develop strategies for reducing wildfire hazard in fire-prone forest landscapes while achieving conservation management goals in other regions.
Author summary
Wildfires represent the most catastrophic disturbance type in North American forests and have caused significant damage and claimed human lives. The implementation of fuel breaks has been considered a strategic measure aimed at reducing fire hazard; however, designing a cost-effective fuel break system is challenging due to unpredictable fire behavior and the uncertain capacity of fuel breaks to stop wildfire spread. Fragmenting the landscape with fuel breaks may also impact other ecosystem services, such as the provision of undisturbed areas to support seasonal wildlife migration. Thus, there is a significant need to develop new methods to assist with the planning and assessment of trade-offs between the capacity of fuel breaks to reduce wildfire hazard versus protecting other ecosystem services. In this study, we applied the fire propagation graph concept to formulate a spatial fuel break planning model that minimizes the spatial connectivity of fuels in a landscape while accounting for potential failures of fuel breaks to contain fire spread. We further applied the Conditional Value-at-Risk concept to address the uncertainty of possible fire spread in fuel break planning decisions. Our model enables a consideration of the trade-off between fuel break width and potential failures to contain fire spread. Our methodology helps develop strategies for reducing wildfire hazard while maintaining key ecosystem services, such as the protection of wildlife migration routes, and could assist in the strategic planning of wildfire mitigation in fire-prone landscapes.
Citation: Yemshanov D, Liu N, Neilson EW, Thompson D, Koch FH, Gritter K (2026) A network optimization framework for fuel break planning in wildfire-prone landscapes: Accounting for fuel break failures and wildlife conservation goals. PLOS Complex Syst 3(8): e0000117. https://doi.org/10.1371/journal.pcsy.0000117
Editor: Víctor M. Eguíluz, IFISC: Instituto de Fisica Interdisciplinar y Sistemas Complejos, SPAIN
Received: September 26, 2025; Accepted: July 8, 2026; Published: August 21, 2026
This is an open access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication.
Data Availability: The Python code to generate the wildfire propagation graphs and relevant model data from the wildfire simulation data is available from https://github.com/nliu-cfs/FPT-plus The relevant input and mapping data and the GAMS model code for problems 1-6 can be found in S4 Appendix.
Funding: Funding for this work was provided by Natural Resource Canada’s Wildfire Risk Management Program (DY, NL, EN, DT). Additional support was provided by the US Department of Agriculture, Forest Service (FHK). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
1. Introduction
Wildfires represent the most frequent natural disturbance type in the montane and boreal forests of North America [1,2] and have caused significant damages to human infrastructure and commercial timber supply [3,4]. Future damages from wildfires and associated mitigation costs are expected to increase due to climate change [5–8], as evidenced by the record-breaking 2023 fire season in Canada [9]. These circumstances call for the implementation of strategic measures to reduce wildfire hazard, such as the removal of forest fuel and disrupting its connectivity across forest landscapes [10–12]. In this context, strategic fuel breaks (cf. [13]), which prescribe the complete removal or reduction of vegetation and fuel, have emerged as a practical means to decrease wildfire intensities to levels where they can be contained [10,11,14,15]. These fuel breaks – strategically placed area treatments or fuel-modified zones – have been implemented in parts of western Canada and USA [16–20].
At a broad scale, these fuel breaks are intended to strategically segment a landscape, thus localizing potential fire damage and reducing the capacity of wildfires to spread over long distances [10–12]. However, designing effective fuel breaks is challenging [15,21,22] because of uncertainties associated with the occurrence of fires and the capacity of fuel breaks to stop fire spread under extreme weather, when firebrand spotting may cause rapid fire spread [23–27]. As a practical matter, fuel breaks must be allocated to individual forest sites, but the combined impact of the fuel breaks on fire hazard must be assessed at a whole-landscape scale. This aspect is difficult to address in planning because wildfire risk shifts in rapidly changing weather, altering which fuel breaks are likely to provide the best mitigation locally and, in turn, the hazard reduction potential of the entire system of fuel breaks.
In Canadian boreal forests, where long-distance firebrand spotting is common, fuel breaks must be sufficiently large to serve as effective barriers [26]. Collectively, a system of fuel breaks also represents a significant landscape-scale disturbance that may impact other ecological processes which rely on forest patch connectivity, such as the movements of wildlife species that require undisturbed habitat to maintain their seasonal behaviour. This is the case for the Southern Mountain population of woodland caribou (Rangifer tarandus caribou; hereafter caribou) in Alberta, Canada. In recent decades, caribou populations have declined due to increased predation caused by anthropogenic fragmentation of caribou habitat [28,29]. Nationwide, woodland caribou is listed as threatened under the Species at Risk Act [30].
Fuel breaks also provide habitat for other ungulate species who forage on early plant regrowth [31] and become more attractive to predators which may also kill caribou. Areas with reduced fuel content provides predators with easier access within habitat areas, further contributing to caribou decline [31–33]. In the West Central region of Alberta, Canada, caribou migrate seasonally from high-elevation areas in summer to low-elevation foothills in winter [34,35], to forage and avoid predators [31,36]. By degrading the ability of caribou to use their preferred migratory corridors, fuel breaks could lead to the loss of sizeable portions of the winter caribou range in this region [37].
Ideally, the fuel breaks, while minimizing the wildfire threat, should pose the minimum impact to caribou migration, but achieving both management goals may not be feasible, particularly in regions where old coniferous forests pose both high burn hazard and remain the primary caribou habitat. In this situation, one needs to assess the trade-off between the anticipated reduction of wildfire hazard versus preserving the key caribou migration routes.
1.1. Assessing the trade-offs between fire hazard reduction and preserving wildlife migration routes
Historically, the planning of fuel treatments and wildlife conservation measures, along with associated assessments of potential trade-offs, have been assisted by optimization. Notable previous applications include the model of León et al. [38] that fragmented contiguous forest areas with fuel while protecting sites with wildlife habitat. The model of Carrasco et al. [12] estimated the impact of firebreak allocation and the reduction of wildfire damage on habitat losses in a landscape. The model of Nguyen et al. [39] aimed to reduce the connectivity of forest fuel and applied a set of simulated fire footprints to allocate firebreaks in a landscape under an uncertain fire regime. The models of Acuna et al. [40] and Troncoso et al. [41] considered timber harvesting as a way to implement fuel breaks and reduce landscape flammability.
Another common fuel treatment planning strategy aims to disrupt local fuel connectivity by controlling the number of adjacent forest patches with fuel [12,42–45]. Extending this idea, several studies have proposed allocations of fuel treatments with the stated goal of reducing the global fuel connectivity across a landscape. In these studies, the landscape was depicted as a network of patches with fuel, and the optimization models were designed to find the locations where fuel breaks would maximally degrade the connectivity of the landscape fuel network [46–49]. Achieving this objective required solving a critical node detection problem [50] that minimized the number of location pairs between which fire spread is possible.
In a related fashion, the concept of patch connectivity has been applied to manage the protection of wildlife corridors [51–53]. Common approaches have included allocating minimum-resistance and minimum-cost wildlife corridors [51,53–56] and maximizing the overall connectivity between patches with habitat in a landscape [12,51,56]. Conrad et al. [56] formulated an allocation of wildlife corridors as a connected graph problem. Similarly, the landscape connectivity problem has also been formulated as a Steiner network problem [55,57] and embedded into broader forest management models featuring other forest planning objectives, such as harvesting [54,58].
To make mitigation effective against large fires such as those that occur in Canadian boreal forests, the capacity of the fuel breaks to stop long-distance fire spread assisted by spotting must be considered explicitly in modelling and planning workflows. This requirement presents a particular challenge to wildlife conservation objectives that has not been fully addressed in previous fuel treatment planning models. In this study, we address this shortcoming and propose a new optimization-based fuel break planning model that aims to reduce wildfire hazard while preserving known wildlife migration routes in the area. We apply the model in a landscape-level setting to find a strategic configuration of fuel breaks that minimizes fire spread connectivity with respect to a large set of uncertain wildfire occurrence outcomes while protecting a target portion of wildlife migration routes in a fire-prone landscape.
1.2. A graph-based approach to depict fire spread uncertainty and fuel break impacts
Fire behaviour in complex landscapes is highly uncertain. A common approach to depict this uncertainty is to simulate a large set of “what-if” scenarios with possible fire footprints and ignition locations within a burn year using a stochastic fire growth model that predicts fire spread as driven by weather, terrain and human activities [59–62]. However, these complex predictions (i.e., tens of thousands of spatial fire scenarios) rarely have been utilized to their full extent to address an important challenge in wildfire mitigation planning: the capacity (or failure) of the planned fuel breaks to stop long-distance fire spread [24,25,63]. A novel aspect of our approach is the conceptualization of a large set of ignition locations and fire footprints that depicts uncertain fire behaviour as a set of directed fire spread graphs, which we use in an optimization setting to find an effective configuration of fuel breaks. Following recent graph theory applications [49,64], we depict the fire spread sequences from ignition locations within the simulated fire footprints as fire propagation graphs (FPGs hereafter). An FPG is an acyclic graph with its root node in the fire’s ignition location and directed edges between adjacent nodes indicating the possible fire spread paths towards its perimeter (Fig 1A). Placing a fuel break within the FPG blocks the fire propagation in that location within the FPG (Fig 1B). Considering that each FPG is a discrete realization of a random fire occurrence and spread process, tracking the impacts of fuel breaks over a large set of equal-probability FPGs estimates the expected burn area (or damage) in the landscape. Compared to previous fuel treatment models based on the FPG concept [49,64] our methodology considers allocation of different fuel break types and can account for potential fuel break failures. Each fuel break type has a defined width that limits the maximum firebrand spotting distance that fuel break type could contain or, otherwise, indicates fuel break failure.
For each fuel break location across the landscape, we estimate the distribution of potential fire spread outcomes as a subset of FPGs overlapping that location. In turn, we characterize each location within each overlapping FPG in terms of burn intensity level using the outputs from the spatial fire growth model. For example, a fuel break of a particular width could stop the spread of a low-intensity fire that is depicted by one FPG but would fail to stop a high-intensity fire that is depicted by another FPG. Tracking fuel break failures in this manner has not been considered previously in a fuel treatment optimization context and helps address a shortcoming of our previous work [47,48] and related studies [64].
The principal objective of our fuel break planning model is to minimize the total burn area. However, we also explored an alternative problem that minimizes the fuel connectivity (and the corresponding fire spread connectivity). Previously, this problem has been formulated as a critical node detection problem that enforces transitive connectivity relationships between nodes in a fuel network [46–48]. Here, we present a new formulation that does not rely on transitive connectivity constraints and instead uses FPGs to track the fire spread connectivity between patches with fuel.
We applied the proposed models to a large forest landscape in West Central Alberta, Canada (Fig 2). In this landscape, we also evaluated the objective of protecting critical caribou migration routes in our fire mitigation scenarios. This associated objective minimizes the number of caribou migration routes fragmented by the fuel breaks. We assessed the trade-off between the degree of caribou migration route protection and the reduction of fire spread potential provided by each allocation approach.
2. Materials and methods
2.1. Study area
The study area is in the Red Rock-Prairie Creek region of Alberta, Canada (Fig 2) and includes upper foothills and lower subalpine ecoregions, with elevations up to 1,750m in the southern portion of the region bordering the Rocky Mountains [70,71]. The region’s forests are dominated by Engelmann spruce (Picea engelmannii) and subalpine fir (Abies lasiocarpa) at higher elevations, and subalpine larch (Larix lyallii) and lodgepole pine (Pinus contorta) at lower elevations. Mixedwood stands at lower elevations are dominated by aspen (Populus tremuloides), white spruce (Picea glauca) and lodgepole pine. The area has been used by forest industry for timber supply [71] and was subject to cyclical oil-and-gas exploration activities that have left a network of narrow linear forest disturbances [72]. The study area partially overlaps the range of the threatened Redrock-Prairie Creek caribou population [28]. During winter, these caribou migrate into our study area from higher elevations in the Rocky Mountains to the southwest in Alberta and British Columbia, where they spend the summer season.
Historically, the area has experienced frequent small to medium-sized fires and infrequent large fires [71]. Fire return intervals are estimated to be between 80 and 300 years at higher elevations in the southern part of the region and between 45 and 476 years in the lower elevation portions of the region [71,73]. The area is characterized by prevailing winds blowing from west-southwest to the east, which influences the dominant fire spread pathways (Fig 3A and 3B). An abundance of mature conifer forests has made the area highly susceptible to fires [71]. In the lower elevation areas, human-caused fires peak in May, while lightning-caused wildfires peak later in the summer. The higher elevation areas in the southwest experience mostly lightning-caused wildfires, with the peak fire season between May and August [74].
2.2. Simulating the fire regime in the study area
We simulated plausible fire regimes in the study area with a spatial fire growth model, which are commonly used for predicting wildfire behaviour in complex landscapes [60–63,75,76]. They have been applied in many regions, for example the Burn-P3 [77] and Cell2Fire [60] models in Canada and the FSim model in the USA [62]. Typically, these models can output location-specific fire behaviour information, such as fire footprints, ignition points, burn intensities and linear rates of spread [61,63,75,77].
We applied the Burn-P3 model [77], which uses the Prometheus fire growth algorithm [78]. The Prometheus algorithm is based on the Canadian Forest Fire Behavior Prediction (FBP) System [79]. Burn-P3 generates a single “burn year” as a stochastic replication. The ignition locations and daily fire weather conditions were based on draws from historical probability distributions for the study area. For each replication of a fire season, the model generated the ignitions and perimeters of individual fires. The fuel composition map was developed in accordance with the Canadian FBP System [80]. The likelihood of ignitions varied across the landscape and was derived from lightning density data [81] or, for human ignitions, based on proximity to roads as calculated from road network data (provided by Weyerhaeuser Ltd.; see details in S1 Appendix). We used 60,000 replications of a burn year at a 1-ha spatial resolution to account for the pattern of fuel and natural fire barriers.
2.3. Fuel break planning with the FPG concept
Our fuel break planning problem considers a forest landscape as a network of N hexagonal patches with fuel (nodes i hereafter, i ∈ N), where each node i is connected to adjacent nodes j by directed edges ij (see Table 1 for symbolic notations). A wildfire can propagate through edges ij from a node with the fire’s ignition location towards the nodes at the fire’s perimeter (Fig 4A and 4B), although the actual fire propagation paths from the ignition location are unknown. We depicted the uncertain behavior of fires in landscape N as a large set F of fire occurrence scenarios. Each scenario f, f ∈ F, depicts the ignition location, the spatial fire footprint and location-specific burn intensity data. We assumed that each fire scenario f has equal probability of occurrence, 1/|F|. The choice of a particular fire growth model is not pivotal to the formulation of our fuel break planning problem, as long as the model can adequately recreate the fire regime of the landscape according to current expectations.
Each fire f is characterized by its spatial footprint Sf and ignition location ϕf. For each location (node) i within the fire’s footprint Sf, the fire growth model outputs provided burn intensity estimates, βfi, in kW-m-1. The fire growth model simulated the burn intensity estimates based on the configuration of fuel types, topography, weather and wind, so the resulting burn intensity pattern within each fire footprint was impacted by all these factors.
For each footprint Sf, we composed the fire propagation graph (FPG) Tf – a directed acyclic graph with the root node ϕf including the fire’s ignition location and directed edges ij connecting adjacent nodes i and j, (i,j)∈Sf, depicting the possible fire propagation paths within the fire’s footprint Sf (Fig 4C). We assumed that fire f could spread to node j from all adjacent upstream nodes i within the fire footprint Sf, such that node j in the FPG Tf is connected to all upstream nodes i. Note that some fire growth models, such as Cell2Fire [60], can generate tree-like graphs that depict the fire propagation to a node from one upstream node only (Fig 1A). Since our study applied the FPGs in the broader context of uncertain high-severity crown fires, where firebrand spotting may create multiple fire spread paths from/to adjacent locations, the FPG must incorporate the uncertain potential of spotting. We depicted the spotting potential in the FPGs by including all plausible ways that a fire could propagate between adjacent upstream and downstream nodes (Fig 1D). This topology better characterizes the uncertainty of local fire behaviour in boreal forests of North America, where crown fires may simultaneously spread from/to multiple directions due to rapid changes of weather, which may occur within hours [26,77,78].
2.3.1. Minimizing the total burn area in the FPG set.
Consider a fire f with a footprint Sf and the corresponding FPG graph Tf. The fire propagation flow starts at the ignition node ϕf to other nodes through the directed edges ij comprising the elements of Tf. Without fuel breaks, the fire is expected to propagate through all edges ij in Tf, (ij)∈Tf. A collection of fuel breaks, if allocated within the fire’s footprint Sf, may stop the spread of fire f through some portions of the FPG and reduce the burned area of fire f from Sf to S′f (Fig 1B and 1D). The burned portion of fire f that persists after allocating the fuel breaks can be found by tracing the fire propagation flow from the ignition node ϕf through the edges ij in Tf that are not blocked by the fuel breaks.
Fires may burn with different intensities, and a decision-maker may choose among different fuel break widths (or types) depending on the anticipated burn conditions in the candidate sites for protection and the available budget. We introduce a set G of possible fuel break types g, g = 1,2,…,|G|, where each type g is characterized by a fuel break width that defines the maximum firebrand spotting distance that this fuel break type can contain in location i. Spotting caused by high-intensity fires is a critical factor of large fire spread [82,83], hence we needed a binary parameter χfig to specify whether the width of fuel break type g is sufficient to contain fire f that burns in node i with intensity level βfi (χfig = 1 and χfig = 0 otherwise). For this study, we derived χfig as follows. We used the local burn intensity level of fire f in node i, βfi (which we generated with the spatial fire growth model), to calculate the maximum firebrand spotting distance for that intensity level using the spotting models of Albini et al. [82], Chase [84] and Thompson et al. [26] (see S2 Appendix). A fuel break type g is expected to contain the spread of fire f through node i (i.e., χfig = 1) if its width is greater than the maximum spotting distance predicted by the spotting model for that location based on the burn intensity level βfi.
The fuel breaks are modelled as the removal of edges between adjacent nodes in landscape N. The binary variable yijg indicates whether a fuel break type g is allocated across edge ij between adjacent nodes i and j, ij ∈ E (yijg = 1, and yijg = 0 otherwise) (Fig 4D). The vicinity area around the removed edge ij where a fuel break segment could be placed is perhaps most easily conceptualized as a rhombus with its longest diagonal equivalent to the edge length between adjacent nodes and shortest diagonal equal to the length of the common boundary between these nodes (Fig 4D). The removal of edge ij is analogous to placing a fuel break segment near the common boundary between nodes i and j (Fig 4Dd). Compared to the depiction of fuel breaks as the removal of nodes – where the fuel break area is defined by the node size [42,47] – the removal of edges enables the selection of a variable fuel break width along the edge between adjacent nodes. The condition when fuel break type g can contain the spread of fire f between nodes i and j can be written as yijgχfigχfjg = 1.
In our formulation, a fire is allowed to propagate to node i from all upstream adjacent nodes in FPG Tf (Fig 1C). The binary decision variable nfi defines the burned status of node i in FTG Tf after fuel break placement (nfi = 1 when fire f spreads to i from the ignition node ϕf). For equal-area nodes i such as those used in this study, the remaining unburned area of fire footprint f after fuel break placement is equivalent to S’f = Σnfi.
The FPGs Tf in fire scenario set F depict the discrete realizations of the uncertain fire regime where each realization f occurs with the probability 1/|F|. Using this setting, we formulate a fuel break allocation problem that minimizes the sum of the remaining burned portions of fire footprints S′f (and corresponding FPGs Tf) over F scenarios in landscape N after allocating the fuel breaks (problem 1 hereafter). Table 2 lists the model equations. Since all fire scenarios f have equal probability of occurrence, minimizing the problem 1 objective (1) is equivalent to minimizing the expected burn area in landscape N. Problem 1 also includes a set of constraints (7) (Eqs.S1.1-S1.3 S1 Appendix) that prevent the creation of loops and clusters, so the fuel breaks appear as linear segments.
Problem 1 minimizes the expected burn area in |F| fire scenarios without penalizing for fuel break failures in some scenarios. In an operational context, high-intensity fires always pose a greater threat than low-intensity fires because of their higher propensity for firebrand production and transport [19] and lower chances of successful containment [85]. A practical fire mitigation plan may include a desired threshold of successful fire containment. Our fuel break allocation problem 2 is based on problem 1 formulation (Eqs.1-7 in Table 2) but adds the constraint (8) that establishes a maximum 5% whole-landscape failure limit for the allocated fuel breaks in landscape N. This limit defines the fire containment probability at the whole-landscape level. It does not enforce the failure threshold for each allocated fuel break segment in the area. In practice, adding a constraint enforcing the 5% failure limit for each fuel break segment would make the solutions more conservative because it would restrict the flexibility to readjust the segment-level fuel break effectiveness while attempting to meet the whole-landscape containment target. Since our application aimed to reduce the wildfire risk across the entire landscape, the use of a whole-landscape threshold felt justified.
Problem 2 prioritizes the reduction of the burned area that may be caused by high-intensity fires and maximizes the capacity of the fuel breaks to stop almost all fires, including those of high severity.
2.3.2. Minimizing the expected worst burn area.
In our problem 1, the uncertainty about which patches might get burned after allocating the fuel breaks is characterized by the distribution of the remaining burned portions of the FPGs over |F| fire scenarios. Minimizing the mean value of that distribution does not guarantee the reduction of the area burned by the largest, most catastrophic fires. Since fire sizes in boreal regions generally follow a power law distribution where small fires dominate [86,87], the largest fires are rare and therefore contribute only modestly to the average burn area over |F| scenarios in objective (1), although their impacts can be disproportionately large in the context of individual scenarios [88,89]. As a practical matter, very large values in the right-hand tail of the burned area distribution may indicate failure to contain the spread of these more destructive fires, which translates to greater damage and higher fire response costs.
When faced with potential failures of the fuel breaks to stop high-intensity fires, a risk-averse decision-maker may consider such failures unacceptable and so will try to minimize these outcomes at all costs. This is an example of tail risk-averse behaviour [90]. A risk-averse decision-maker evaluates potential actions in terms of the minimum utility that might emerge from selecting these actions. If confidence about estimating the potential fire spread outcomes is poor, a tail risk-averse strategy is to prioritize the best of the expected worst outcomes.
Minimizing the expected worst burn area requires minimizing the expected value of the right tail of the distribution of the burn area across |F| scenarios. This can be achieved by minimizing an upper-percentile-based metric, such as Conditional Value-at-Risk (CVaR) [91–93]. In our case, CVaR for a confidence level α, α∈[0;1], can be defined as the expected burn area after fuel break placement over the (1–α)×100% worst scenarios (Fig 5). For a random variable, CVaRα is the conditional mean of the values (i.e., the remaining burn area after fuel break placement, Σnfi) exceeding VaRα, which is the value in the distribution of burn areas that is exceeded only in (1–α)×100% of the scenarios. We set the α level to 0.95. This level represents the point when the size of an upper tail of the fire size distribution is sufficient to track the most threatening large fire events without making it too sensitive to the occurrence of the extreme cases in a discrete set F of simulated fire scenarios (see S1 Appendix).
The CVaR is convex with respect to the decision variables [92,93]. For discrete scenario-based distributions such as our problems 1 and 2, minimizing CVaR can be expressed by a set of linear equations [92]. Analogously to Vilches et al. [64], our problem 3 depicts the tail risk-averse strategy of avoiding the expected worst-case burn outcomes by minimizing the CVaR of the burn area. For a distribution of |F| scenarios with occurrence probabilities 1/|F|, problem 3 with CVaRα minimization is approximated with the objective function (9) and equivalent sets of |F| + 1 auxiliary decision variables and |F| + 1 inequality constraints (10) (Table 2). The objective (9) minimizes the expected worst burned area over |F| fire scenarios after allocating the fuel breaks. Term in constraint (10) in Table 2 defines the burned area after fuel break placement in fire scenario f, and symbols Z and µf define the auxiliary decision variables.
Problem 4 that minimizes the expected worst-case burn area with a 5% fuel break failure limit is formulated analogously to problem 3 but with added constraint (10).
2.3.3. Minimizing the fire spread connectivity.
An alternative approach to reduce wildfire hazard is to reduce the connectivity between patches with fuel by minimizing the total number of node pairs between which fire spread is possible. For each node pair ij in landscape N, one must estimate whether the allocated fuel breaks could disrupt fire propagation from node i to node j in one or more fire scenarios f. Previously, the allocation of fuel treatments to minimize fuel connectivity was solved as a critical node detection problem [46–48], which finds the key nodes in a fuel network whose removal maximally degrades its connectivity. However, this formulation did not permit selection of alternative fuel break types, nor did it account for potential fuel break failures. We addressed this shortcoming with our planning problem 5 whose objective minimizes the number of node pairs between which fire spread is possible in one or more fire scenarios f (Table 2). For each pair of ignition node i and destination node j in landscape N, the binary decision variable uij in objective (11) in Table 2 specifies whether a fire could spread from i to j in one or more fire scenarios f (i.e., uij = 1). For each node i, set Ωi in Equations (11,12) defines a fireplain that delineates the area where wildfires could spread from i during a designated timespan (cf. [94]) (Fig 6). The right side of Equation (12) denotes the sum of the burn status variables nfj at destination node j for all fires f that were ignited in origin node i and able to spread to node j. Set Fi in Eq. (12) includes the fires f with their ignition locations in node i. The right side of Equation (12) becomes positive if there is a connecting flow path from node i to node j through the FPGs in one or more fire scenarios f after the allocation of fuel breaks.
Problem 6, which minimizes the number of node pairs with possible fire spread under a 5% fuel break failure limit, is formulated analogously to problem 5 but with added constraint (8). Table 3 lists the model scenarios.
2.4. Protecting the wildlife migration routes in the fire mitigation scenarios
Forest landscapes may include critical corridors which are used by wildlife (such as caribou) as a part of their seasonal behaviour [95]. Mapping the movements of these animals through a landscape is done by tracking a sample of GPS-collared individuals. Their movements are depicted as a collection of paths crossing the landscape, with migratory movements being represented by the long path lengths [95,96]. For this case study, we used GPS collars on 189 caribou collared between 1998 and 2016 from the Red Rock-Prairie Creek and Narraway caribou herds to identify a set of migration routes that overlap our study area. We have converted GPS relocations to connected paths and removed the paths with missing data which produced unrealistically long paths. We extracted the start and end points of the top 1% longest path segments and converted them to straight line features which yielded 422 straight line features from 155 caribou individuals. We only used the resulting migration routes that overlapped our study area, as such, most routes came from the Red Rock-Prairie Creek data. We then subset the resulting line features most frequently aligned with the northeast-southwest directions connecting the winter and summer ranges of the two herds. Using these we defined a set R of caribou migration routes r, r ∈ R, each representing a segmented path digitized from caribou movement data at the spatial resolution of network nodes i (Fig 7C). A path r can overlap one or more nodes i and is characterized by length ρr, measured as the total number of nodes i it overlaps. The path length ρr serves as a path conservation priority measure, with the longest paths receiving the highest priority.
Background map shows the burn probability; d) caribou migration route protection priorities for edges ij. Background map shows the expected burn area; e) proportion of each node with west-, south- and southwest-facing slopes facing the prevailing winds, used to identify suitable sites for fuel break placement; f) edges ij representing potential candidates for fuel break placement. The map data can be found in S4 Appendix. Mapping software used: Python Matplotlib 3.4.
A binary parameter, σrij, indicates whether path r overlaps the area around edge ij in landscape N where a fuel break segment may be established (σrij = 1 when path r intersects either node i or j or both). We assume that an overlap occurs when path r crosses any part of nodes i or j. This assumption accounts for the distance (at least 500 m) at which the negative impact of disturbances on caribou populations remains significant [97,98].
The allocation of fuel breaks is often most effective in regions with large amounts of old coniferous forest, which tends to be highly flammable. In our study area, these conifer forests also have good capacity to support caribou populations, so there is a trade-off between minimizing the fire spread potential in areas with conifer stands vs. protecting as many caribou migrations routes in those areas as possible. Since the fuel breaks had to be wide (i.e., ≥ 360 m) to be effective in the presence of firebrand spotting, the creation of such large disturbance spaces would significantly disrupt caribou movement behavior. We treated the disruptions as binary events and introduced a binary variable, qr, that tracks whether caribou path r crosses one or more areas near edges ij with allocated fuel breaks. For caribou path r, constraint (13) specifies qr = 1 when the path is fragmented by fuel breaks and constraint (14) specifies qr = 0 when r has no intersecting fuel breaks, i.e.,:
A decision-maker defines the maximum proportion of the caribou migration routes, qmax that can be intersected by fuel breaks. Constraint (15) ensures that the proportion of the caribou paths disrupted by the fuel breaks does not exceed qmax., i.e.,:
2.5. Building fire propagation graphs from wildfire simulation data
We defined the fuel breaks (i.e., strategic fuel barriers) as spatial units with removed or substantially reduced forest fuel that can be established as a part of standard forestry operations and whose size would be compatible with the size of large clearcuts. We set the node size in fuel network N depicting the study area to 400 ha, which was sufficient to track major fire behaviour patterns and the caribou migration routes in the area. This node size limited the maximum fuel break width that could be evaluated in problems 1–6 to the cross-sectional width of a node (2.15 km), which is equivalent to the length of edge ij between a node i and an adjacent node j (see Fig 4D). We adopted a large node size to account for the possibility of spotting caused by high-intensity fires under extreme weather [99]. Ideally, a fuel break should be of sufficient width to contain potential spotting for most predicted fires. Using the maximum spotting distance models of Albini et al. [82], we estimated that a 2.15-km node cross-section would exceed the estimated maximum spotting distances for 99.5% of the predicted fires in the study area with fire line intensities up to 70,000 kW/m; fuel break widths below this 2.15-km maximum would be expected to contain a progressively smaller subset of predicted fires. Thus, our assessment of fuel break effectiveness was based on a relatively conservative underlying assumption. Since our fuel break planning was intended to assist with strategic mitigation of wildfire risk, having wide fuel breaks was deemed important for our study area because – due to its remote location – fire crews’ response times are expected to be long, especially during the peak fire season. This conservative approach is also consistent with regional management decisions, presented in the draft Upper Smoky Sub-Regional Plan [100], that anticipate the creation of wide strategic fire barriers in the area. Within each simulated fuel break, less flammable fuels, such as broadleaf trees and more open vegetation were assumed, which allows for standard fire suppression activities using fixed-wind air tankers [85] that are commonly used in fire suppression in Alberta.
We built the fire propagation graphs Tf using ignition locations and fire footprints Sf simulated for the study area (see S1 Appendix for the procedure of composing the FPGs). Parts of the study area with proportions of south-, west- and southwest-facing slopes (i.e., that face the prevailing winds) equal to 20% or higher were set as suitable candidates for fuel break placement (Fig 7E and 7F). We considered five fuel break types g with corresponding widths of 0.36, 0.67, 1.1, 1.55 and 2.15 km. For each fuel break type g, we calculated the maximum burn intensity level this fuel break type could potentially contain using the spotting distance model of Albini et al. [82] (see S2 Appendix).
The maximum corresponding fire intensity levels in node i that could be contained by the fuel break types g were estimated as 3500, 6500, 15000, 32000 and 70500kW-m-1, respectively. For each node i in fire footprint Sf, we set the values of the parameter χfig – which specified whether fuel break type g could stop fire f in node i – by comparing the burn intensity levels βfi estimated with the fire growth model in i with the corresponding intensity threshold values calculated with the spotting distance model for a particular width of fuel break type g. S1 Appendix provides the model solution procedure.
3. Results
3.1. Mapping the fuel break solutions
The fuel network comprising the study area included 834 nodes and 4788 edges (Fig 7A and 7B). The area is dominated by mature conifer forests and, as noted earlier, is highly susceptible to fires driven by prevailing winds from west-southwest (Fig 3A). The fire spread directions and the footprints of the simulated wildfires were influenced by the prevailing wind pattern (Fig 3B), with 95% of fires expected to spread over distances of 12.9 km or less. The northern and eastern portions of the study area located at lower elevations are characterized by warmer weather, the occurrence of both human-caused and lighting-caused fires and higher projected burn risk than the areas located at higher elevations in the southwestern region of the area (Fig 7C), which tend to experience colder weather and mostly lighting-caused fires.
The maps in Figs 8 and 9 show the fuel break solutions for the fuel break areas of 100 and 150 km2. The background maps in Figs 8 and 9 indicate the expected burn area in node i after fuel break placement. The fuel breaks tended to be allocated as linear segments that were roughly perpendicular to the directions of the prevailing winds (Fig 3A).
The background maps show the number of FPGs with burned status after fuel break allocation. The 5% fuel break failure threshold is applied at the whole-landscape level. The map data can be found in S4 Appendix. Mapping software used: Python Matplotlib 3.4.
The background maps show the number of FPGs with burned status after fuel break allocation. The 5% fuel break failure threshold is applied at the whole-landscape level. The map data can be found in S4 Appendix. Mapping software used: Python Matplotlib 3.4.
3.2. Minimizing the expected burn area vs. fuel connectivity
Problem 1 solutions, which minimized the expected burn area, generally segmented the landscape by means of moderate-length fuel breaks (Figs 8A and 9A). In conditions with a strong directional wind pattern such as in our study (see Fig 3A), the impacts of fuel breaks are likely to have limited depth. In our case, the most feasible strategy is to segment the forest landscape by evenly spaced fuel breaks oriented perpendicularly to the prevailing winds. The fuel breaks were rarely placed in high-elevation areas in the southwestern portion of the landscape where colder weather is less conducive to the spread of large fires than in the low-elevation areas. To maximize the total length of the allocated fuel breaks, problem 1 tended to allocate lowest-width (360 m) fuel breaks (Figs 8A and 9A), which resulted in a high (30.9-35.6%) fuel break failure rate (Table 4, Fig 10). (See Figs S3.1 and S3.2 in S3 Appendix depicting the fuel break failure rates as pink haloes around the allocated segments). The high failure rate stems from the formulation of the problem 1 objective, which minimizes the expected burn area over F fire scenarios but does not minimize the burn area of the largest or most intense fires. Nevertheless, a policy of maximizing the total fuel break length at the cost of their lower overall effectiveness is feasible for the study area, because the predicted fire size distribution is dominated by small, moderate-severity fires (which could be contained by moderate-width fuel breaks) (Figs 8A and 9A).
X-axis denotes the total fuel break area, B, in km2. Performance metric: a) expected burn area, no fuel break failure limit; b) expected burn area, 5% fuel break failure limit; c) expected worst burn area, no fuel break failure limit; d) expected worst burn area, 5% fuel break failure limit; e) the number of node pairs with possible fire spread, no fuel break failure limit; f) the number of node pairs with possible fire spread, 5% fuel break failure limit; g, h) average fuel break failure rate with and without the fuel break failure constraint. Lower values on the y-axis indicate better outcomes.
Adding the 5% fuel break failure threshold in problem 2 solutions created wider fuel breaks (Table 4) aimed at containing most of the high-severity fires in the landscape driven by spotting (Figs 8B and 9B). Most fuel breaks were positioned in low-elevation regions, and the total fuel break length was shorter than in problem 1 solutions (Figs 8B and 9B vs. Figs 8A and 9A).
In problem 3 solutions, which minimized the expected worst burn area, some fuel breaks were allocated in high-elevation regions with lower fire hazard (Figs 8C and 9C). Because the chance of large fire occurrence, while rare, existed across the study area, the allocation of fuel breaks was more diffuse than in problem 1 solutions. The average fuel break failure rate was lower than the failure rate in problem 1 solutions but still within the 25.9-33.5% range (Table 4). Note that problem 3 aimed to reduce the fire hazard in only those scenarios with the largest burn areas. Since the fire distribution in the area was dominated by small fires, minimizing the occurrence of only the largest fires targeted a limited subset of the potential fires in the area, which explains the poorer combined performance of problem 3 solutions versus problem 1 solutions (Table 4).
Adding the 5% fuel break failure threshold in problem 4 solutions resulted in fuel break allocations similar to problem 2 solutions (Figs 8D and 9D, Table 4). Typically, the allocated fuel breaks were wider than in problem 3 solutions, in some cases at the maximum node width of 2.15 km.
In problem 5 solutions that minimized fuel connectivity, the fuel breaks were allocated in longer segments aimed to partition the landscape (Figs 8E and 9E). Problem 5 solutions had a lower fuel break failure rate than problems 1 and 3 solutions (Table 4). Adding the 5% fuel break failure limit in problem 6 solutions resulted in a fuel break pattern that resembled problems 2 and 4 solutions; however, the fuel breaks were arranged in fewer but longer segments. Most of the fuel breaks were allocated in segments 1.1 km to 1.55 km wide (Table 4).
3.3. Fuel break performance vs. the allocated treatment area
As the total fuel break area increased, the optimal solutions increasingly segmented the landscape into partial compartments (Figs 8 and 9). In problems 1, 3 and 5 without the fuel break failure threshold, reducing the expected burn area and fire spread connectivity followed the rule of diminishing returns (Figs 10A, 10C and 10E): larger-budget solutions had lower marginal effectiveness. Problem 1, 3 and 5 solutions performed differently in dimensions of the expected burn area and the fire spread potential. Problem 5 solutions achieved the most effective reduction of the fire spread potential across all budget levels (Fig 10E), whereas problem 1 solutions were the most effective in reducing the expected burn area (Fig 10A). Problems 1, 3 and 5 performed similarly in reducing the expected worst burn area (Fig 10C). Problems 2, 4 and 6 with the 5% fuel break failure threshold were unable to reduce the expected burn area as effectively as problem 1 solutions (Fig 10B), but their capacity to reduce the wildfire spread connectivity was close to problem 5 solutions without the fuel break failure limit (Fig 5F). Problem 3 that minimized the expected worst burn area performed closely to problems 2, 4 and 6 (Fig 10D).
Maintaining a low fuel break failure rate by keeping them wide imposed a moderate penalty on the capacity to reduce the overall fire spread potential (Fig 10F). The net reduction of the expected burn area in problems 2, 4, and 6 solutions was 25% lower than in problem 1 solutions that achieved the best overall reduction of the expected burn area (Fig 10B callout I). The trade-off between the reduction of the expected burn area (problem 1 objective) and minimizing the fire spread connectivity (problem 5 objective) was moderate (Fig 11 callout I). Notably, the solutions minimizing the expected worst burn area (problems 3 and 4) performed poorly in both dimensions. Comparatively poorer performance was also exhibited by the solutions with a 5% fuel break failure limit, but the performance penalty was relatively small in absolute terms (Fig 11). Thus, adding a fuel break failure limit could be beneficial for reducing the risk of spread of large fires with little penalty on the reduction of the expected burn area.
Callout I shows the trade-off between the problem 5 solution that minimizes the fire spread potential and the problem 1 solution that minimizes the expected burn area.
3.4. Minimizing the wildfire spread potential and the protection of caribou migration routes
Protecting the caribou migration routes had a moderate impact on the reduction of fire spread risk connectivity (Fig 12, Table 5) because the model tended to allocate the fuel breaks in low-elevation areas in the eastern and northern parts of the landscape and avoided the high-elevation areas in the southwestern regions with the greatest concentration of caribou migration routes (Figs 8, 9, 13 and 14). Notable disruption of caribou migration routes was only observed in large-budget solutions to problems 3–6 (Fig 12, Table 5). The overall impact of adding the caribou migration route protection requirements on solutions minimizing the expected burn area (problems 1 and 2) was modest (Table 5). This is because the hotspots with high expected burn area tended to occur in lower-elevation areas whereas caribou migration routes did not penetrate too deeply into low-elevation regions (Fig 7C and 7D). Notably, the solutions to problems 3–6 that allocated a more dispersed fuel break pattern showed higher sensitivity to the caribou migration route protection requirements.
The background maps show the number of FPGs with burned status after fuel break allocation. The 5% fuel break failure threshold is applied at the whole-landscape level. The map data can be found in S4 Appendix. Mapping software used: Python Matplotlib 3.4.
The background maps show the number of FPGs with burned status after fuel break allocation. The 5% fuel break failure threshold is applied at the whole-landscape level. The map data can be found in S4 Appendix. Mapping software used: Python Matplotlib 3.4.
4. Discussion
The proposed FPG-based approach builds on our previous work that optimized fuel treatments [47,48] in the following ways. Foremost, the new model enables selection among different fuel break types. The width of each fuel break type defines its capacity to stop the spread of a fire that burns at a particular intensity level βfi, which was used to estimate the maximum spotting distance from that location. Previous studies have demonstrated that the negative effects of wildfires, such as fire intensity and damage to vegetation, can be mitigated within fuel treatment boundaries [101–104]. In this context, the capacity to select a particular fuel break width makes the fuel break design better tailored to local fire spread conditions.
Another notable aspect of the proposed approach is the adoption of a scenario-based formulation where each candidate location for a fuel break is characterized by a distribution of potential fire spread outcomes. In some scenarios, the chosen fuel break width may be insufficient to contain high-intensity fires; by considering all fire scenarios it is possible to estimate the expected fuel break failure rates in those locations.
Fuel break planning is typically undertaken at the scale of individual forest sites, but the uncertain consequences of these local decisions cause impacts at a broader, landscape scale. Our approach estimates these landscape-scale impacts by tracking the fire propagation flow in a set of FPGs depicting the possible fire spread conditions in an area of interest. Note that the impact of fuel break failures could be significant if the area is predicted to experience intense fires. This makes the FPG-based approach particularly useful in regions where large fires are predicted to occur, such as boreal and montane regions of Canada, as evidenced by the record-breaking 2023 fire season [9] as well as recent wildfires in Alberta, including fires in Fort McMurray in 2016 and Jasper in 2024. Segmenting a fire-prone landscape to minimize fuel connectivity via application of problems 5 or 6 can be a workable strategy to reduce the potential risk of spread of large fires [48]. In fact, there has been some adoption of the landscape partitioning concept in part of western Canada (e.g., [105,106]), generally following the concept of Potential Wildfire Operational Delineations (PODs) adopted in the USA [107]. The FPG-based approach, thanks to its capacity to track the burn area, could help with the development of such PODs-like fire mitigation measures. In principle, our single-period model can be applied for long-term strategic planning by applying it in sequence or reformulation to a multi-period planning problem.
In boreal and montane forests of Canada, the concept of 1–2-km wide fuel breaks is becoming increasingly relevant at the landscape scale [19,20,100]. To be practical, fuel break allocation must be planned as a part of other activities which tend to create large human disturbances. In these forests, clearcut harvesting is the most common activity that creates large disturbances with reduced fuel content. This makes harvesting a feasible option for implementation of the strategic fuel break concept. The most immediate way to establish large fuel breaks is to incorporate their planning into an operational harvest planning process. Such an approach would need to evaluate the effectiveness of harvesting in terms of both reducing wildfire hazard and maximizing harvest revenue. For practical implementation, one would need to link the problem 1–6 formulations with an optimization-based harvest scheduling problem [108]. Such an integrated model would help design harvesting sequences that achieve timber supply objectives while reducing the threat of wildfires and potential timber losses.
4.1. Minimizing the expected burn damage vs. fire spread connectivity
Our results point to notable differences between mitigation strategies that reduce the expected burn area versus fuel connectivity per se. Minimizing fuel connectivity does not guarantee the reduction of the average burn area because fuel connectivity and the locations of hotspots with high burn risk do not always correlate. For example, patches with fuel can be well-connected to facilitate fire spread but the overall susceptibility to fire could be low. Such conditions characterize the high-elevation regions in the southwestern portion of our study area, where colder, less fire-conducive weather caused an estimated lower burn probability than in the low-elevation regions (Fig 7F).
Adding a 5% fuel break failure limit imposed a stabilizing effect on the fuel break pattern. For example, in the scenarios with total fuel break area B = 150 km2, 35.4% of the fuel break segments in solutions to problems 2,4 and 6 appeared in similar locations. In contrast, only 10% of the allocated segments were placed in similar locations in problems 1,3 and 5 solutions without a fuel break failure rate limit.
4.2. Accounting for the fire behaviour uncertainty
Our study articulates the importance of an adequate representation of the fire behaviour uncertainty space in fire mitigation planning. We depicted the multitude of fire spread outcomes in a landscape as a large set of fire scenarios with unique fire footprints. However, a finite set of fire outcomes may not fully capture the whole uncertainty space of fire behaviours and could limit the model applicability in extreme weather conditions that are not captured by the scenario set. This emphasizes the need for accurate parameterization of the fire growth model, so the outputs can provide an adequate characterization and sampling of the uncertainty space. Notably, the issue of undervaluing the uncertainty of local fire behaviour was partially addressed by considering the worst-case local fire spread outcomes through fully connected FPGs that allowed a fire to spread to a location of concern from any adjacent upstream location. This assumption implies that the manager acknowledges the uncertainty of local fire spread and plans for the worst-case outcome. While this configuration felt sufficient to track the possibility of firebrand spotting conditions, more complex algorithms may be required to build FPGs in regions where spotting is rare or spatially constrained by local factors. Updating fire growth models to better characterize local fire spread conditions and provide a more comprehensive depiction of possible fire behaviors in the target landscape could be a valuable improvement of current fire modelling tools.
Our optimization model used a set F of FPGs generated exogenously with a spatial fire growth model. In the prediction of fire spread and severity, assumptions about ignition probabilities and weather impact the outputs of the fire growth model. This emphasizes the importance of proper calibration of the fire model and an understanding of its sensitivities to changes in the weather and ignitions inputs. Notably, the Burn-P3 model used in this study has a track record of previous applications and sensitivity analyses in Canadian forests [109–111]. Potentially, other fire growth models could be used to generate the fire scenarios. In this context, a multi-model comparison of sensitivities to weather and ignition patterns and their impacts on the fuel break configurations could be an interesting methodological exercise. Furthermore, predicted weather data, such as the daily fire weather projections presented in [89,112,113], could be adapted to simulate future fire regimes and develop fire mitigation strategies under climate change. This could be an important area of research given that climate change is expected to increase the frequency and severity of future fires in Canadian forests [114].
4.3. Impact of the caribou migration route protection requirements
The trade-off between the protection of caribou migration routes and the reduction of fire hazard was driven by the need to protect potential hotspots with large amounts of fuel with fuel breaks, while also protecting the critical caribou migration routes in those locations. Our results show a moderate impact of caribou migration route protection on the capacity of fuel breaks to reduce the wildfire spread potential. The disrupting effect of fuel breaks on caribou migration routes depended on the degree of spatial overlap between the hotspots with high burn risk (i.e., where the protective impact of fuel breaks would be most noticeable) and the locations of the caribou migration routes. In our case, this overlap was limited. However, more research into how fuel break placement may compromise the amount, connectivity and quality of caribou habitat specific to their winter range, rather than the migration routes is needed. In the Red Rock region, a decrease in caribou herd migration to their winter range is attributed to increased forest harvesting and oil and gas development [29,35,37]. Consequently, even if the migration routes are protected, their use by caribou may still decrease further if fuel breaks are located elsewhere in their winter range.
Although fuel breaks may cause a loss of caribou habitat, wildfires may do the same. Recently burned forests are likely avoided by caribou [115] and may have more forage for other ungulates that attract predators [116,117]. Because climate change is expected to lead to more frequent wildfires in the region [114], the exposure of caribou migration routes to large wildfires is expected to increase accordingly. As a compromise approach for maximizing the protection of caribou migration routes without substantially increasing the fire spread potential, one could consider less impactful fuel treatments in the migration route space, such as conversion to less flammable forest types [118]. Nevertheless, a forest patch burned by a wildfire may not be as detrimental to caribou as the same area removed by harvest [119,120], so protecting forest habitat from wildfire via forest harvest must be considered carefully. As such, comparing the habitat losses caused by wildfires with the losses caused by firebreak placement could help assess the rationale for building firebreaks in a caribou range area and could be an interesting topic of further research.
The firebreak locations prioritized by our models do not present catastrophic loss of the known caribou migration routes in the study area. However, the association of the observed decline in winter migration by the Red Rock caribou herd with increasing habitat disturbance [29,35,37] suggests that any further removal of forest through harvest will further negatively impact caribou populations viability. At a minimum, the increase of early seral habitat after the creation of fuel breaks provides movement and hunting opportunities for caribou predators [120]. Consequently, the fuel break solutions identified in our study will require additional research investigating how fuel treatments may compromise the amount, connectivity and quality of winter habitat in the study area. In a broader context, our proposed methodology could help elucidate the trade-offs of building fuel breaks in other caribou ranges, and whether additional anthropogenic disturbances might render those areas uninhabitable by caribou [29,35,37,120,121]. In turn, this could help develop wildfire mitigation options in other landscapes where large fires, human development and wildlife populations coexist.
Supporting information
S1 Appendix. Generating the Fire Propagation Graphs (FPGs) Tf.
https://doi.org/10.1371/journal.pcsy.0000117.s001
(PDF)
S2 Appendix. The firebrand spotting distance calculation.
https://doi.org/10.1371/journal.pcsy.0000117.s002
(PDF)
S3 Appendix. Fuel break allocation solutions with a depiction of the fuel break failure rates.
https://doi.org/10.1371/journal.pcsy.0000117.s003
(PDF)
Acknowledgments
The authors express gratitude and special thanks to Wendy Crosina and Lyle Dechief from Weyerhaeuser Inc. for their assistance with the geospatial data characterizing the study area. This work was completed as part of a long-standing research collaboration between Natural Resources Canada, Canadian Forest Service and the US Department of Agriculture, Forest Service on topics pertaining to forest health and management.
References
- 1. Dennison PE, Brewer SC, Arnold JD, Moritz MA. Large wildfire trends in the western United States, 1984–2011. Geophys Res Lett. 2014;41(8):2928–33.
- 2. Hanes CC, Wang X, Jain P, Parisien M-A, Little JM, Flannigan MD. Fire-regime changes in Canada over the last half century. Can J For Res. 2019;49(3):256–69.
- 3. Theobald DM, Romme WH. Expansion of the US wildland–urban interface. Landsc Urban Plann. 2007;83(4):340–54.
- 4.
Thomas DS, Butry DT, Gilbert SW, Webb DH, Fung JF. The Costs and Losses of Wildfires. Spec. Publ. NIST SP-1215; 2017.
- 5. Delfino RJ, Brummel S, Wu J, Stern H, Ostro B, Lipsett M, et al. The relationship of respiratory and cardiovascular hospital admissions to the southern California wildfires of 2003. Occup Environ Med. 2009;66(3):189–97. pmid:19017694
- 6. Dennekamp M, Abramson MJ. The effects of bushfire smoke on respiratory health. Respirology. 2011;16(2):198–209. pmid:20920143
- 7. Abatzoglou JT, Battisti DS, Williams AP, Hansen WD, Harvey BJ, Kolden CA. Projected increases in western US forest fire despite growing fuel constraints. Commun Earth Environ. 2021;2(1).
- 8. Ellis TM, Bowman DMJS, Jain P, Flannigan MD, Williamson GJ. Global increase in wildfire risk due to climate-driven declines in fuel moisture. Glob Chang Biol. 2022;28(4):1544–59. pmid:34800319
- 9. Jain P, Barber QE, Taylor SW, Whitman E, Castellanos Acuna D, Boulanger Y, et al. Drivers and impacts of the record-breaking 2023 wildfire season in Canada. Nat Commun. 2024;15(1):6764. pmid:39164286
- 10. Alcasena FJ, Ager AA, Salis M, Day MA, Vega-Garcia C. Optimizing prescribed fire allocation for managing fire risk in central Catalonia. Sci Total Environ. 2018;621:872–85. pmid:29216595
- 11. Ager AA, Evers CR, Day MA, Alcasena FJ, Houtman R. Planning for future fire: Scenario analysis of an accelerated fuel reduction plan for the western United States. Landsc Urban Plann. 2021;215:104212.
- 12. Carrasco J, Mahaluf R, Lisón F, Pais C, Miranda A, de la Barra F, et al. A firebreak placement model for optimizing biodiversity protection at landscape scale. J Environ Manage. 2023;342:118087. pmid:37196613
- 13. Ascoli D, Russo L, Giannino F, Siettos C, Moreira F. Firebreak and fuelbreak. In: Manzello S, editor. Encyclopedia of wildfires and wildland-urban interface (WUI) fires. Cham: Springer. 2018.
- 14. Agee JK, Bahro B, Finney MA, Omi PN, Sapsis DB, Skinner CN, et al. The use of shaded fuelbreaks in landscape fire management. For Ecol Manage. 2000;127(1–3):55–66.
- 15. Ager AA, Day MA, Aparício BA, Houtman R, Stinchfield A. Optimizing the implementation of a forest fuel break network. PLoS One. 2023;18(12):e0295392. pmid:38091301
- 16. Finney MA. Design of regular landscape fuel treatment patterns for modifying fire growth and behavior. For Sci. 2001;47(2):219–28.
- 17. Schmidt DA, Taylor AH, Skinner CN. The influence of fuels treatment and landscape arrangement on simulated fire behavior, Southern Cascade range, California. For Ecol Manage. 2008;255(8–9):3170–84.
- 18. Tubbesing CL, Fry DL, Roller GB, Collins BM, Fedorova VA, Stephens SL, et al. Strategically placed landscape fuel treatments decrease fire severity and promote recovery in the northern Sierra Nevada. For Ecol Manage. 2019;436:45–55.
- 19. Marshall G, Thompson D, Anderson K, Simpson B, Linn R, Schroeder D. The impact of fuel treatments on wildfire behavior in North American Boreal Fuels: a simulation study using FIRETEC. Fire. 2020;3(2):18.
- 20. North MP, Bisbing SM, Hankins DL, Hessburg PF, Hurteau MD, Kobziar LN, et al. Strategic fire zones are essential to wildfire risk reduction in the Western United States. fire ecol. 2024;20(1).
- 21. Collins BM, Stephens SL, Moghaddas JJ, Battles J. Challenges and approaches in planning fuel treatments across fire-excluded forested landscapes. J For. 2010;108(1):24–31.
- 22. Rönnqvist M, D’Amours S, Weintraub A, Jofre A, Gunn E, Haight RG, et al. Operations research challenges in forestry: 33 open problems. Ann Oper Res. 2015.
- 23.
Christiansen EC. Fuel treatments in the 21st century – do they matter? In: Keane RE, Jolly M, Parsons R, Riley K, editors. Proceedings of the large wildland fires conference; May 19-23, 2014; Missoula, MT. Proc. RMRS-P-73. Fort Collins, CO: U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station; 2015. pp. 267–8.
- 24. Kalies EL, Yocom Kent LL. Tamm review: are fuel treatments effective at achieving ecological and social objectives? A systematic review. For Ecol Manage. 2016;375:84–95.
- 25. Syphard AD, Keeley JE, Brennan TJ. Factors affecting fuel break effectiveness in the control of large fires on the Los Padres National Forest, California. Int J Wildland Fire. 2011;20(6):764–75.
- 26. Thompson DK, Yip DA, Koo E, Linn R, Marshall G, Refai R, et al. Quantifying firebrand production and transport using the acoustic analysis of in-fire cameras. Fire Technol. 2022;58(3):1617–38.
- 27.
Baxter G, Richards M, Tymstra C. A wildland-urban post-fire case study: The Grouse Complex B.C. 2023. Institute for Catastrophic Loss Reduction (ICLR) research paper series 77. Toronto, ON; 2024. Available from: https://www.iclr.org/wp-content/uploads/2024/11/Grouse-Complex-WU-Fires-N.pdf
- 28.
Environment Canada (EC). Recovery Strategy for the Woodland Caribou, Southern Mountain population (Rangifer tarandus caribou) in Canada [Proposed]. Species at Risk Act Recovery Strategy Series. Ottawa: Environment Canada; 2014. pp. viii + 68:vi + 55.
- 29. Williams SH, Steenweg R, Hegel T, Russell M, Hervieux D, Hebblewhite M. Habitat loss on seasonal migratory range imperils an endangered ungulate. Ecol Sol and Evidence. 2021;2(1).
- 30.
Species at Risk Act (SARA). Bill C-5, An act respecting the protection of wildlife species at risk in Canada. 2002 Available from: http://laws.justice.gc.ca/PDF/Statute/S/S-15.3.pdf
- 31. McKay TL, Finnegan LA. Predator–prey co‐occurrence in harvest blocks: Implications for caribou and forestry. Conserv Sci Prac. 2022;4(12).
- 32. Dickie M, Serrouya R, McNay RS, Boutin S. Faster and farther: wolf movement on linear features and implications for hunting behaviour. J Appl Ecol. 2016;54(1):253–63.
- 33. Dickie M, McNay SR, Sutherland GD, Cody M, Avgar T. Corridors or risk? Movement along, and use of, linear features varies predictably among large mammal predator and prey species. J Anim Ecol. 2020;89(2):623–34. pmid:31648375
- 34. McDevitt AD, Mariani S, Hebblewhite M, Decesare NJ, Morgantini L, Seip D, et al. Survival in the Rockies of an endangered hybrid swarm from diverged caribou (Rangifer tarandus) lineages. Mol Ecol. 2009;18(4):665–79. pmid:19175508
- 35.
Lamb C, Steenweg R, Serrouya R, Hervieux D, M c n a y S, Heard D, et al. The erosion of threatened mountain caribou migration. Biol Conserv. 2025:1–16.
- 36. Apps CD, McLellan BN, Kinley TA, Flaa JP. Scale-dependent habitat selection by Mountain Caribou, Columbia Mountains, British Columbia. J Wildlife Manage. 2001;65(1):65.
- 37. MacNearney D, Pigeon K, Stenhouse G, Nijland W, Coops NC, Finnegan L. Heading for the hills? Evaluating spatial distribution of woodland caribou in response to a growing anthropogenic disturbance footprint. Ecol Evol. 2016;6(18):6484–509. pmid:27777724
- 38. León J, Reijnders VMJJ, Hearne JW, Ozlen M, Reinke KJ. A landscape-scale optimisation model to break the hazardous fuel continuum while maintaining habitat quality. Environ Model Assess. 2018;24(4):369–79.
- 39. Nguyen D, Wei Y, Belval EJ, Thompson MP, Gannon BM, Young JD, et al. An optimization model to prioritize fuel treatments within a landscape fuel break network. PLoS One. 2024;19(12):e0313591. pmid:39689080
- 40. Acuna MA, Palma CD, Cui W, Martell DL, Weintraub A. Integrated spatial fire and forest management planning. Can J For Res. 2010;40(12):2370–83.
- 41. Troncoso JJ, Weintraub A, Martell DL. Development of a threat index to manage timber production on flammable forest landscapes subject to spatial harvest constraints. INFOR: Inf Syst Oper Res. 2016;54(3):262–81.
- 42. Pais C, Carrasco J, Elimbi Moudio P, Shen Z-JM. Downstream protection value: Detecting critical zones for effective fuel-treatment under wildfire risk. Comput Oper Res. 2021;131:105252.
- 43. Rachmawati R, Ozlen M, Reinke KJ, Hearne JW. An optimisation approach for fuel treatment planning to break the connectivity of high-risk regions. For Ecol Manage. 2016;368:94–104.
- 44. Rachmawati R, Ozlen M, Hearne J, Reinke K. Fuel treatment planning: fragmenting high fuel load areas while maintaining availability and connectivity of faunal habitat. Appl Math Modell. 2018;54:298–310.
- 45. Russo L, Russo P, Siettos CI. A complex network theory approach for the spatial distribution of fire breaks in heterogeneous forest landscapes for the control of wildland fires. PLoS One. 2016;11(10):e0163226. pmid:27780249
- 46. Matsypura D, Prokopyev OA, Zahar A. Wildfire fuel management: Network-based models and optimization of prescribed burning. Eur J Oper Res. 2018;264(2):774–96.
- 47. Yemshanov D, Liu N, Thompson DK, Parisien M-A, Barber QE, Koch FH, et al. Detecting critical nodes in forest landscape networks to reduce wildfire spread. PLoS One. 2021;16(10):e0258060. pmid:34618859
- 48. Yemshanov D, Liu N, Neilson EW, Koch FH, Parisien M-A. Evaluating fuelbreak strategies for compartmentalizing a fire-prone forest landscape in Alberta, Canada. PLoS One. 2025;20(5):e0321722. pmid:40397881
- 49. Yemshanov D, Liu N, Neilson EW, Thompson D, Koch FH. A graph-based optimization framework for firebreak planning in wildfire-prone landscapes. Ecol Inform. 2025;90:103339.
- 50. Veremyev A, Prokopyev OA, Pasiliao EL. An integer programming framework for critical elements detection in graphs. J Comb Optim. 2014;28(1):233–73.
- 51. Dilkina B, Houtman R, Gomes CP, Montgomery CA, McKelvey KS, Kendall K, et al. Trade‐offs and efficiencies in optimal budget‐constrained multispecies corridor networks. Conserv Biol. 2016;31(1):192–202.
- 52. St John R, Tóth SF, Zabinsky ZB. Optimizing the geometry of wildlife corridors in conservation reserve design. Oper Res. 2018;66(6):1471–85.
- 53. Yemshanov D, Dawe DA, Bakalarczyk A, Liu N, Boulanger Y, Boucher J, et al. Balancing wildlife protection and wildfire threat mitigation using a network optimization approach. Front For Glob Change. 2023;6.
- 54. Yemshanov D, Haight RG, Liu N, Rempel R, Koch FH, Rodgers A. Exploring the tradeoffs among forest planning, roads and wildlife corridors: a new approach. Optim Lett. 2021;16(3):747–88.
- 55.
Lai KJ, Gomes CP, Schwartz MK, McKelvey KS, Calkin DE, Montgomery CA. The Steiner multigraph problem: wildlife corridor design for multiple species. In: Proceedings of the Twenty-Fifth AAAI Conference on Artificial Intelligence (AAAI-11), August 7-11, 2011. San Francisco, CA, USA; 2011.
- 56. Conrad JM, Gomes CP, van Hoeve W-J, Sabharwal A, Suter JF. Wildlife corridors as a connected subgraph problem. J Environ Econ Manage. 2012;63(1):1–18.
- 57. Sessions J. Solving for habitat connections as a steiner network problem. For Sci. 1992;38(1):203–7.
- 58. Carvajal R, Constantino M, Goycoolea M, Vielma JP, Weintraub A. Imposing Connectivity constraints in forest planning models. Oper Res. 2013;61(4):824–36.
- 59. Wei Y, Rideout D, Kirsch A. An optimization model for locating fuel treatments across a landscape to reduce expected fire losses. Can J For Res. 2008;38(4):868–77.
- 60. Pais C, Carrasco J, Martell DL, Weintraub A, Woodruff DL. Cell2Fire: a cell-based forest fire growth model to support strategic landscape management planning. Front For Glob Change. 2021;4.
- 61. Riley K, Thompson M, Scott J, Gilbertson-Day J. A model-based framework to evaluate alternative wildfire suppression strategies. Resources. 2018;7(1):4.
- 62. Finney MA, McHugh CW, Grenfell IC, Riley KL, Short KC. A simulation of probabilistic wildfire risk components for the continental United States. Stoch Environ Res Risk Assess. 2011;25(7):973–1000.
- 63. Parisien M-A, Junor DR, Kafka VG. Comparing landscape-based decision rules for placement of fuel treatments in the boreal mixedwood of western Canada. Int J Wildland Fire. 2007;16(6):664–72.
- 64. Vilches M, Carrasco J, Dávila S, Palacios D, Quezada F, Weintraub A. A two-stage stochastic programming approach incorporating spatially-explicit fire scenarios for optimal firebreak placement. Eur J Oper Res. 2024.
- 65.
2021 Canadian Census of Population. Administrative boundaries. Available from: https://www12.statcan.gc.ca/census-recensement/2021/geo/sip-pis/boundary-limites/index2021-eng.cfm?year=21
- 66.
Canadian National Burned Area Composite. A spatial dataset. 2024. Available from: https://cwfis.cfs.nrcan.gc.ca/datamart/metadata/nbac
- 67.
Canadian National Fire Database. 2024. Available from: https://cwfis.cfs.nrcan.gc.ca/ha/nfdb
- 68.
Alberta’s Base Waterbody Polygon Update. 2023. Available from: https://open.alberta.ca/opendata/gda-475e5500-dfd5-4835-9adb-a1f80c183b5c
- 69.
Alberta Vegetation Inventory. Available from https://open.alberta.ca/opendata/gda-3dbcfa02-e97a-4059-9414-1ed8e0700e80
- 70.
Government of Alberta, Natural Regions Committee. Natural Regions and Subregions of Alberta. In: Downing DJ, Pettapiece WW. Government of Alberta; 2006.
- 71.
TECO Natural Resources Group Limited. Weyerhaeuser Grande Prairie 2011 – 2021 DFMP Landscape Assessment FMA #6900016. Prepared for Weyerhaeuser Company Limited by TECO Natural Resource Group Limited. Edmonton. AB; 2011. Available from: https://www1.agriculture.alberta.ca/$Department/deptdocs.nsf/all/formain15767/$file/Appendix-1-LandscapeAssessment.pdf?OpenElement
- 72. Finnegan L, MacNearney D, Pigeon KE. Divergent patterns of understory forage growth after seismic line exploration: Implications for caribou habitat restoration. For Ecol Manage. 2018;409:634–52.
- 73.
Andison D. Landscape-level fire activity on foothills and mountain landscapes of Alberta. Alberta Foothills Disturbance Ecology Research Series, Report No. 2. Hinton, AB: Foothills Model Forest; 2000.
- 74.
Tymstra C, Wang D, Rogeau M-P. Alberta wildfire regime analysis. Alberta Sustainable Resource Development, Forest Protection Division, Wildfire Policy and Business Planning Branch. Wildfire Science and Technology Report PFFC-01-05; 2005.
- 75. Parisien M-A, Dawe DA, Miller C, Stockdale CA, Armitage OB. Applications of simulation-based burn probability modelling: a review. Int J Wildland Fire. 2019;28(12):913.
- 76. Miller C, Ager AA. A review of recent advances in risk analysis for wildfire management. Int J Wildland Fire. 2012;22(1):1–14.
- 77.
Parisien MA, Kafka V, Hirsch KG, Todd JB, Lavoie SG, Maczek PD. Mapping Wildfire Susceptibility with the BURN-P3 Simulation Model. Edmonton, AB: Natural Resources Canada, Canadian Forest Service, Northern Forestry Centre, Information Report NOR-X-405; 2005.
- 78.
Tymstra C, Bryce RW, Wotton BM, Taylor SW, Armitage OB. Development and structure of Prometheus: The Canadian wildland fire growth simulation model. Edmonton, AB: Natural Resources Canada, Canadian Forest Service, Northern Forestry Centre, Information Report NOR-X-417. 2010. https://www.fs.usda.gov/rm/pubs/rmrs_p073.pdf#page=206
- 79. Stocks BJ, Lynham TJ, Lawson BD, Alexander ME, Wagner CEV, McAlpine RS, et al. Canadian forest fire danger rating system: an overview. For Chron. 1989;65(4):258–65.
- 80.
Forestry Canada Fire Danger Group. Development and structure of the Canadian Forest Fire Behavior Prediction System. Ottawa, ON: Forestry Canada Information Report ST-X-3; 1992.
- 81.
Environment and Climate Change Canada. Lightning Density Data; 2023. Available from: https://open.canada.ca/data/en/dataset/75dfb8cb-9efc-4c15-bcb5-7562f89517ce
- 82. Albini FA, Alexander ME, Cruz MG. A mathematical model for predicting the maximum potential spotting distance from a crown fire. Int J Wildland Fire. 2012;21(5):609–27.
- 83. Manzello SL, Suzuki S, Gollner MJ, Fernandez-Pello AC. Role of firebrand combustion in large outdoor fire spread. Prog Energy Combust Sci. 2020;76. pmid:32116406
- 84.
Chase CH. Spot fire distance equations for pocket calculators. Research Note No. INT-310. USDA Forest Service, Intermountain Forest and Range Experiment Station; 1981. Available from: https://www.frames.gov/catalog/38784
- 85. Wheatley M, Wotton BM, Woolford DG, Martell DL, Johnston JM. Modelling initial attack success on forest fires suppressed by air attack in the province of Ontario, Canada. Int J Wildland Fire. 2022;31(8):774–85.
- 86. Stocks BJ, Mason JA, Todd JB, Bosch EM, Wotton BM, Amiro BD, et al. Large forest fires in Canada, 1959-1997. J Geophys Res. 2003;108(D1):8149.
- 87. Malamud BD, Millington JDA, Perry GLW. Characterizing wildfire regimes in the United States. Proc Natl Acad Sci U S A. 2005;102(13):4694–9. pmid:15781868
- 88. Metsaranta JM. Potentially limited detectability of short-term changes in boreal fire regimes: a simulation study. Int J Wildland Fire. 2010;19(8):1140–6.
- 89. Wang X, Parisien M-A, Taylor SW, Perrakis DDB, Little J, Flannigan MD. Future burn probability in south-central British Columbia. Int J Wildland Fire. 2016;25(2):200–12.
- 90. Gilboa I, Schmeidler D. Maxmin expected utility with non-unique prior. J Math Econ. 1989;18(2):141–53.
- 91. Acerbi C, Tasche D. Expected shortfall: a natural coherent alternative to value at risk. Econ Notes. 2002;31(2):379–88.
- 92. Rockafellar RT, Uryasev S. Optimization of conditional value-at-risk. J Risk. 2000;2:21–41.
- 93. Rockafellar RT, Uryasev S. Conditional value-at-risk for general loss distributions. J Banking Fin. 2002;26(7):1443–71.
- 94.
Scott JH, Thompson MP. Emerging concepts in wildfire risk assessment and management. In: Keane RE, Jolly M, Parsons R, Riley, editors. 2015. Proceedings of the large wildland fires conference; May 19-23, 2014; Missoula, MT. Proc. RMRS-P-73. Fort Collins, CO: U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station; 2015. pp. 196–206.
- 95. Ferguson SH, Elkie PC. Seasonal movement patterns of woodland caribou (Rangifer tarandus caribou). J Zool. 2004;262(2):125–34.
- 96. Rudolph TD, Drapeau P. Using movement behaviour to define biological seasons for woodland caribou. Ran. 2012;:295–307.
- 97.
Environment Canada. Scientific Assessment to Inform the Identification of Critical Habitat for Woodland Caribou (Rangifer tarandus caribou), Boreal Population, in Canada: 2011 Update. Ottawa, Canada: 2011.
- 98.
Environment Canada. Recovery Strategy for the Woodland Caribou (Rangifer tarandus caribou), Boreal population, in Canada. Species at Risk Act Recovery Strategy Series. Environment Canada, Ottawa, ON. 2012. Available from: http://www.registrelep-sararegistry.gc.ca/virtual_sara/files/plans/rs%5Fcaribou%5Fboreal%5Fcaribou%5F0912%5Fe1%2Epdf
- 99. Wadhwani R, Sullivan C, Wickramasinghe A, Kyng M, Khan N, Moinuddin K. A review of firebrand studies on generation and transport. Fire Saf J. 2022;134:103674.
- 100.
Government of Alberta. Upper Smoky Sub-Regional Plan. 2025. [cited 2025 Aug 29] Available from: https://www.alberta.ca/system/files/epa-draft-sub-regional-plan-upper-smoky-plan.pdf
- 101. Safford HD, Stevens JT, Merriam K, Meyer MD, Latimer AM. Fuel treatment effectiveness in California yellow pine and mixed conifer forests. For Ecol Manage. 2012;274:17–28.
- 102. Parks SA, Miller C, Holsinger LM, Baggett LS, Bird BJ. Wildland fire limits subsequent fire occurrence. Int J Wildland Fire. 2015;25(2):182–90.
- 103. McKinney ST, Abrahamson I, Jain T, Anderson N. A systematic review of empirical evidence for landscape-level fuel treatment effectiveness. Fire Ecol. 2022;18(1).
- 104. Hunter ME, Iniguez JM, Lentile LB. Short- and long-term effects on fuels, forest structure, and wildfire potential from prescribed fire and resource benefit fire in Southwestern Forests, USA. Fire Ecol. 2011;7(3):108–21.
- 105.
Forest Practices Board. Forest and Fire Management in BC: Toward Landscape Resilience. Special Report SR51. 2023. Available from: https://www.bcfpb.ca/wp-content/uploads/2023/06/SR61-Landscape-Fire-Management.pdf
- 106.
Impact Assessment Agency of Canada. Canadian Impact Assessment Registry. Lake Louise Community Fire Guard. 2024. Available from: https://iaac-aeic.gc.ca/050/evaluations/proj/81249
- 107. Buettner WC, Beeton TA, Schultz CA, Caggiano MD, Greiner MS. Using PODs to integrate fire and fuels planning. Int J Wildland Fire. 2023;32(12):1704–10.
- 108. Johnson KN, Scheurman HL. Techniques for prescribing optimal timber harvest and investment under different objectives - discussion and synthesis. For Sci. 1977;23(Suppl. S1):a0001–z0001.
- 109. Parisien M-A, Parks SA, Miller C, Krawchuk MA, Heathcott M, Moritz MA. Contributions of ignitions, fuels, and weather to the spatial patterns of burn probability of a boreal landscape. Ecosystems. 2011;14(7):1141–55.
- 110. Parks SA, Parisien M-A, Miller C. Multi-scale evaluation of the environmental controls on burn probability in a southern Sierra Nevada landscape. Int J Wildland Fire. 2011;20(7):815–28.
- 111. Parisien MA, Dawe DA, Miller C, Stockdale CA, Armitage OB. Applications of simulation-based burn probability modelling: a review. Int J Wild Fire. 2019;28:913–26.
- 112. Stralberg D, Wang X, Parisien M, Robinne F, Sólymos P, Mahon CL, et al. Wildfire‐mediated vegetation change in boreal forests of Alberta, Canada. Ecosphere. 2018;9(3).
- 113. Wang X, Studens K, Parisien M-A, Taylor SW, Candau J-N, Boulanger Y, et al. Projected changes in fire size from daily spread potential in Canada over the 21st century. Environ Res Lett. 2020;15(10):104048.
- 114. Boulanger Y, Gauthier S, Burton PJ. A refinement of models projecting future Canadian fire regimes using homogeneous fire regime zones. Can J For Res. 2014;44(4):365–76.
- 115. Konkolics S, Dickie M, Serrouya R, Hervieux D, Boutin S. A burning question: what are the implications of forest fires for Woodland Caribou? J Wildl Manag. 2021;85(8):1685–98.
- 116. Collins WB, Dale BW, Adams LG, Mcelwain DE, Joly K. Fire, grazing history, lichen abundance, and winter distribution of caribou in Alaska’s taiga. J Wildl Manag. 2011;75(2):369–77.
- 117. Peters W, Hebblewhite M, Smith KG, Webb SM, Webb N, Russell M, et al. Contrasting aerial moose population estimation methods and evaluating sightability in west‐central Alberta, Canada. Wildlife Soc Bull. 2014;38(3):639–49.
- 118. Ager AA, Vaillant NM, Finney MA. A comparison of landscape fuel treatment strategies to mitigate wildland fire risk in the urban interface and preserve old forest structure. For Ecol Manage. 2010;259(8):1556–70.
- 119. DeMars CA, Serrouya R, Mumma MA, Gillingham MP, McNay RS, Boutin S. Moose, caribou, and fire: have we got it right yet? Can J Zool. 2019;97(10):866–79.
- 120. Stevenson S, Finnegan L, Johnson C, McKay T. Differential responses of woodland caribou to fire and forestry across boreal and montane ecosystems—a literature review. For: Int J For Res. 2024;98(4):478–93.
- 121. Wittmer HU, Sinclair ARE, McLellan BN. The role of predation in the decline and extirpation of woodland caribou. Oecologia. 2005;144(2):257–67. pmid:15891849