1 Introduction

Wildfires spread along trajectories set by a coincident wind direction. Wind accelerates fire spread by pushing hot gasses through the flaming fire front, igniting fuel in advance of the fire (Linn 2019). Wind speed and direction are site-specific and variable such that a single daily observation at a distant weather station may provide little insight into fire spread and behavior (e.g., Amiro et al. 2004; Podur and Wotton 2011). Analysis of local observations of wind speeds during wildfire-induced evacuations (Beverly and Bothwell 2011; Beverly et al. 2011) has shown that extreme winds are an important factor driving wildfire threats to public safety. The importance of wind speed and direction in precipitating wildfire disasters in western Canada cannot be overstated. In 2011, southeast winds gusting over 80 km h−1 drove a fire into the Town of Slave Lake, Alberta, and surrounding communities, destroying over 480 homes (Flat Top Complex Wildfire Review Committee 2012). The 2016 Horse River Fire destroyed over 2500 dwellings in Fort McMurray when winds shifted to the southwest and strong gusts of up to 40 km h−1 drove the fire into the community (MNP 2017; KPMG 2017). Likewise, the 2020 fire that destroyed Lytton, British Columbia, was pushed into the village by observed southerly winds of 35 km h−1 with gusts of 50 km h−1 or greater (Cohen and Westhaver 2022). During these disasters, the fire entered the built environment by traveling along predominantly stable wind-driven trajectories, traversing distances of 1.7–3 km in a single hour (MNP 2017; Wildfire Science Documentation Group 2012). Under the most extreme wind conditions, such as those observed during the 1968 Vega Fire, wind-driven fires in western Canada have traveled at rates of 6 km h−1, covering distances of tens of kilometers in a single run (Flat Top Complex Wildfire Review Committee 2012).

Despite the highly directional nature of wildfire threats to public safety, landscape fire risk assessments are typically omnidirectional. Monte Carlo fire growth simulations can be used to estimate burn probabilities at a given location (e.g., Salis et al. 2021), but these burn probabilities alone provide no information about which directions the simulated fires traveled to reach that location. Likewise, statistical modeling methods (e.g., Macauley et al. 2022) and machine learning-based approaches (e.g., Tonini et al 2020) used to map fire-susceptible locations typically consider environmental attributes such as land cover and proximity to roads or water bodies, but ignore the configuration or directionality of such features.

Simulation modeling has been used to describe major fire-flow paths across a landscape (e.g., Alcasena et al. 2017), and circuit-theoretic models have been used to assess landscape connectivity with respect to fire spread (e.g., Gray and Dickson 2015). Mahmoud and Chulahwat (2020) estimated the most probable fire paths from a graph network model to calculate the mean probability of a fire reaching a community from a particular wind direction. Price et al. (2015) used empirical statistical modeling to estimate the probability of a fire spreading to a point of interest along sample lines drawn from 20 possible ignition points located randomly at distances ranging from 400 m to 22 km. Price and Bedward (2020) further developed the approach to include ignition probabilities.

Regardless of whether they are omnidirectional or directional in nature, all probabilistic approaches to wildfire risk assessment combine static and dynamic environmental factors that influence when and where wildfires burn. Static factors such as the locations of hazardous fuels or the presence of non-flammable areas such as lakes can be known with relative certainty and will remain predominantly unchanged over both long and short decision horizons. Whereas, dynamic factors such as fuel moisture, wind direction, and ignition locations vary over short durations and small spatial extents (e.g., Hilton et al. 2015; Quill et al. 2019; Penman et al. 2020), and cannot be predicted at a specified location with any degree of certainty more than a few days in advance, if at all (e.g., Van Wagner 1987, Wotton and Martell 2005, Taylor et al. 2013). Macauley et al. (2022) used this distinction in decision horizon length to justify modeling static factors independently when mapping spatial variation in fire spread potential for strategic planning purposes.

Decoupling static and dynamic fire environment variables in landscape fire risk assessments was also advocated by Beverly et al. (2021) as an alternative to computationally complex simulation models. In that study, a simple deterministic metric was developed to rate the potential for fire transmission to a location based solely on proximity to nearby hazardous fuels. Despite ignoring topography (i.e., slope and aspect) and stochastic fire controls (i.e., ignitions and weather), the exposure metric exhibited better alignment with real wildfires in Alberta, Canada, than complex burn probability simulations (i.e., Beverly and McLoughlin 2019). Exposure is a physical quantity measured for each cell in a landscape grid to rate the potential for fire transmission over relatively localized ignition distances (e.g., 0–500 m). As with burn probabilities, exposure alone offers no insight about the direction of fire transmission; however, by examining the amount, configuration and orientation of high-exposure areas distributed around a locale of interest, such as a community, the potential for fire encroachment from outlying areas becomes apparent (e.g., Fig. v, Beverly et al. 2021).

In this study, we develop a simple, standardized method for measuring and displaying the directional vulnerability of a locale, or receptor, based on the pattern of high-exposure areas present in the surrounding landscape. Directional environmental data can be conceptualized with a circular form around a central location, where points on the surrounding perimeter correspond to compass directions. A common example is the rose or polar diagram, a form of radial graph used to display quantities of a variable in a circular distribution, such as the frequency of observed wind directions (Rautenhaus et al. 2018). Here, we develop a customized rose diagram for displaying directional properties of land-based data; in our case, the distribution of high-exposure areas along 15-km directional trajectories connecting outlying areas to a central receptor (i.e., community). First, we analyze historical burned areas to document the pattern of underlying high-exposure areas associated with observed continuous wildfire trajectories in the province of Alberta, Canada. Results are used to define a viable fire trajectory based on the portion of a trajectory’s length that intersects high exposure. We then compute the presence or absence of these viable trajectories for 5-km segments (inner, middle, and outer) of each 15-km trajectory plotted for 360 possible wind directions around a given community. We apply the method to several hundred small- to moderately-sized communities in Alberta. Results are summarized graphically to describe directional vulnerabilities of point features and compiled across management units to describe regional vulnerabilities. Ongoing research and potential applications of the directional assessments are discussed, including: prepositioning and prioritization of limited fire suppression resources, planning fuel reduction treatments, proactively identifying candidate locations for operational activities, assessing potential transportation vulnerabilities during wildfire evacuations, and scenario planning.

2 Methods

2.1 Study area

A provincial geoadministrative dataset (Government of Alberta 2021) was used to identify 1503 small- to moderately-sized communities in Alberta. For the purposes of this study, a community refers to a concentration of residents and built structures. Dataset records included a classification of the community type, which was defined predominantly by population size thresholds ranging from small hamlets with as few as five dwellings to cities with a population greater than 10,000 (Table 1). We restricted our analysis to community types and sizes conducive to representation with a single point location (i.e., centroid), which we considered incompatible with the roughly 20 cities located in the province.

Table 1 Categories used to describe communities in Alberta, Canada (Municipal Services and Legislation Division of Alberta Municipal Affairs, n.d.)
Full size table

For each community, we defined a 70,686-ha circular assessment area with a 15-km radius from the community centroid. Community centroids were used in lieu of irregular administrative boundaries to standardize the area under consideration and facilitate a systematic analysis of hundreds of communities across the entire province, which is consistent with other coarse-scale studies of community-wildfire interactions (e.g., Dye et al. 2021). Circular buffers around each community were limited to 15 km based on wildfire evacuation records compiled by Beverly and Bothwell (2011). In that study, the distance of the fire from the community at the time of the evacuation was documented for 158 evacuation events, and in 89% of these cases, this distance was less than 15 km.

2.2 Data sources and processing

Thematic land cover maps in a raster format with a 100 m × 100 m resolution were obtained for the years 2016, 2019 and 2021 from the Alberta Wildfire Management Branch. These datasets are maintained by the province of Alberta for the purpose of wildfire modeling. Areas disturbed by forest harvesting and wildfire are updated annually using detailed mandatory cutblock reports from forest management companies and delineated wildfire perimeters. Landscape fire exposure was calculated following the methods detailed by Beverly et al. (2021). The process involved transforming non-fuel land cover classes and assigned fuel types of the Canadian Forest Fire Behavior Prediction (FBP) System (Forestry Canada Fire Danger Group 1992) into a binary hazardous fuel layer. All forested fuel types composed of at least 25% coniferous tree species such as black spruce (Picea mariana (Mill.) BSP) and lodgepole pine (Pinus contorta Dougl. ex Loud. var. latifolia) were considered hazardous fuels capable of transmitting fire over distances of 500 m and were assigned a value of 1. Remaining cells were assigned a value of 0 and consisted of water, non-fuel areas such as roads and recently burned areas as well as deciduous stands and open cover types such as grass. Exposure to hazardous fuels was calculated for each cell in the raster as the percent of cells in a circular neighborhood with a 5-cell (500 m) radius that was classified as hazardous fuels.

A simplified raster denoting high-exposure areas was produced by reclassifying percent exposure into two categories: Cells with a minimum of 60% exposure were deemed high exposure, and those with less than 60% exposure were deemed low exposure. The 60% high-exposure threshold follows the findings of Beverly et al. (2021) that the majority of burned areas in Alberta occur in locations with exposure ≥ 60%. Insight about the extent of overlap expected between a viable fire trajectory and high-exposure areas was obtained by combining 2016 and 2019 exposure rasters with fire polygons mapped in the 3 years following the exposure assessment.

A database of 690 historical wildfire polygons documenting the perimeters of fires that occurred in the province of Alberta from 2016 to 2021, inclusive was obtained from the Alberta Wildfire Management Branch. Fire polygons ranged in size from less than 1 ha to over 400,000 ha and were mapped with a variety of methods including ground and airborne Global Positioning Systems (GPS), infrared scanning, digitized aerial photographs, satellite imagery classification, and hand sketches. Some polygons generated from hand sketches or non-corrected ground or airborne GPS lacked sufficient detail for our analysis and were omitted. Likewise, small fires that burned less than 100 ha were omitted to ensure sample trajectories were of a length consistent with the previously described major fire runs associated with wildfire disasters in western Canada. The anomalous 2019 Chuckegg Creek Fire was omitted from analysis due to prior reports of it burning predominantly in a non-hazardous fuel type (i.e., trembling aspen [Populus tremuloides]) under extreme fire weather conditions coupled with long-term drought (MNP 2020; Beverly et al. 2021). In total, 95 fires were retained for analysis, most of which had detailed perimeters generated from satellite imagery or digitized from aerial photos.

2.3 Defining a viable fire trajectory

Sample trajectories within historical fire polygons were delineated manually. Fire polygons typically consist of one or more lobe-shaped areas that are consistent with elliptical fire growth along a stable trajectory (i.e., Van Wagner 1969). Parallel or duplicate trajectories were avoided by delineating a single central trajectory within the core area of a given fire lobe. The following criteria were used to ensure the sampled trajectories were independent and reasonably representative of continuous major fire runs that pose a threat to populated areas: Each linear trajectory spanned a minimum of 1000 m wholly within the mapped burned area of the fire; no unburned patches or fire refugia were intersected by the trajectory; all portions of the trajectory were at least 50 m from the mapped fire’s edge; and multiple trajectories within a single fire did not overlap. An example fire perimeter and associated sample trajectories are shown in Fig. 1A. Due to size and shape constraints, not all of the 95 fire polygons analyzed could accommodate at least one trajectory. In total, 573 trajectories were delineated within the boundaries of 79 mapped wildfires. The length of sample trajectories ranged from 1013 m to 17,415 m, with an average length of 2489 m. Collectively, the sampled trajectories spanned 1426 km, with 47 individual trajectories exceeding 5 km in length.

Fig. 1
figure 1

A Sample transects positioned within the burned area of an example fire perimeter (HWF-177-2018). B Sample transects for fire HWF-177-2018 in relation to high-exposure areas shown in gray and low-exposure areas shown in white. C Total number of sample transects (n = 573) by proportion intersection with high-exposure areas

Full size image

Sample trajectories were overlaid on the binary high-exposure layer (Fig. 1B) to measure the length of each trajectory that intersected areas of high exposure. The vast majority of sample trajectories exhibited extensive overlap with high-exposure areas (Fig. 1C). On average, sample trajectories within burned areas intersected high exposure for 79% of their length. Median intersection with high-exposure areas was 99%. For the purposes of our directional analysis, we, therefore, defined a viable fire trajectory as one with at least 80% of its length traversing high exposure (i.e., ≥ 60% exposure). This viability criterion was met by almost three-quarters (n = 404) of our sampled transects.

2.4 Measuring directional vulnerability

Below, we provide a step-by-step description of our methodological process using a hypothetical example assessment point (Fig. 2), which could be a community, asset, or any location of value. These same steps were repeated for each of our community centroids using the 2021 land cover raster. All geoprocessing was conducted in ArcGIS Pro (Esri Inc. 2021).

Fig. 2
figure 2

Land cover surrounding an example assessment point (e.g., a value, asset, or community). Fuel types of the Canadian Forest Fire Behavior Prediction (FBP) System are shown along with water and non-fuel land cover classes. Circular buffers at 5, 10, and 15 km are shown in black

Full size image

Landscape fire exposure in the circular assessment area (Fig. 3A) and a simplified high-exposure binary layer (Fig. 3B) were calculated following the methods detailed in Sect. 2.2. Two example directional trajectories are displayed as arrows in Fig. 3B to highlight that some directions coincide with continuous high-exposure areas, while others do not. For any given trajectory, the extent of intersection with underlying high exposure can vary by distance interval (e.g., 0–5 km, 5–10 km, and 10–15 km), as illustrated by the example trajectory at 0° (due north).

Fig. 3
figure 3

A Landscape fire exposure surrounding an example assessment point computed from the land cover grid shown in Fig. 2, following the methods presented in Beverly et al. (2021). Circular buffers at 5, 10, and 15 km are shown in black. B Shaded areas indicate high-exposure values (i.e., ≥ 60%). Arrows denote example directional trajectories: The northerly trajectory exhibits little overlap with high-exposure areas, beyond the 0–5-km buffer distance; whereas, the westerly trajectory exhibits continuous overlap with high-exposure areas at all buffer distances

Full size image

We delineated 360 possible wildfire trajectories around the assessment point at 1° directional intervals (Fig. 4). Circular buffers positioned at 5-km intervals were used to divide each directional trajectory into three equal-length segments for analysis: 0–5 km, 5–10 km, and 10–15 km. The length of each directional trajectory segment that intersected high exposure was computed for all 1080 directional segments.

Fig. 4
figure 4

Plot of 1° directional trajectories (n = 360) emanating from the example assessment point (e.g., a value, asset, and community). Colors denote 5-km segments that form the length of each 15-km trajectory: red, 0–5 km; orange, 5–10 km; and yellow, 10–15 km. High- and low-exposure areas from Fig. 3B underlay the directional trajectories

Full size image

The viability of each directional trajectory segment was evaluated using the threshold we derived from our analysis of sample trajectories within historical fire polygons in Sect. 2.3: A viable fire trajectory was considered one with at least 80% of its length traversing high exposure (i.e., ≥ 60% exposure). We used the resulting binary viability classification (i.e., viable and not viable) to produce a graphical visualization and quantitative summary of the distribution of high-exposure pathways around the assessment point. To display results in an intuitive format, we designed a customized rose or polar diagram that used shading to indicate the presence or absence of a viable trajectory segment within a given spatially referenced distance class (Fig. 5A).

Fig. 5
figure 5

A Directional trajectories that exhibited alignment with high-exposure areas (i.e., 80% of the trajectory length intersected high exposure), by 5-km segment. B Directional trajectories that exhibited high-exposure alignment for all three distance segments (0–5 km, 5–10 km, and 10–15 km)

Full size image

Our custom rosette is novel and fundamentally different from a standard rose or polar graph in which the y-axis displays a quantity being measured in relation to a direction. The most common usage of rose diagrams for environmental data is the wind rose (Rautenhaus et al. 2018), which displays the frequency that an observed wind direction was recorded at a given weather station during a specified time period. In our rosette, the y-axis is not used to measure a frequency or quantity, rather it displays spatially referenced distances extending outwards from the assessment point: 0–5 km, 5–10 km, and 10–15 km. The presence of a viable trajectory within each distance class (i.e., inner, middle, and outer extents) is indicated by color-coded shading. The result is a highly intuitive, spatially referenced visualization of the distribution of viable fire pathways at each 1° directional interval around a point of interest, as well as an indication of the continuity of those pathways over the full length of each 15-km directional trajectory. We produce a further simplified version of the graph by showing only those directional pathways that have continuous viable trajectories for all three distance bands (Fig. 5B).

2.5 Community and regional applications

We expect the magnitude and distribution of directional vulnerability to wildfire to differ not only by community, but also by region. Some communities will be vulnerable to wildfire encroachment from a particular wind direction, while others will not. To assess directional vulnerability associated with individual communities, the methods outlined in Sect. 2.4 were applied to several hundred small- and medium-sized communities in Alberta. Circular assessment areas surrounding community centroids that had little to no areas assessed as having high exposure were not expected to generate any viable trajectories. To reduce the computational load of our calculations and focus on the most vulnerable communities, we analyzed communities that had at least 20% of their circular assessment area classified as high exposure.

Our use of centroids to represent communities meant that assessments of viable trajectories at 0–5 km from centroid could be affected by the size of the community and the extent to which developed lands presented as non-fuels within the inner 5-km buffer around the community centroid. While this is not an issue when assessing individual communities, it means that compiling collective summaries for multiple communities of different sizes could introduce inconsistencies. To summarize results irrespective of community size, we, therefore, focused on the 5–15-km distance from community centroid.

At a regional scale, multiple communities within broader administrative zones could exhibit high loads of viable wildfire trajectories aligned with particular wind directions. To produce regional summaries across broad administrative units, the collective frequency or load of viable directional pathways was compiled for all communities within a given Forest Area. Forest Areas are administrative units used by the province’s wildfire management branch and number ten in total. For these regional summaries, we used traditional rose diagrams that displayed the frequency of continuous viable trajectories over the 5–15-km distance class for a given 1° direction, summed for all communities within the Forest Area.

3 Results

Of the 1503 communities investigated, 590 had high exposure in less than 20% of their circular assessment area and were, therefore, omitted from further analysis. Most of these omitted communities were located outside the provincial Forest Protection Area (FPA) in areas classified as the “white zone,” which contains all of the province’s privately-owned land and the majority of agricultural lands (Ruan et al. 2016). The remaining 913 communities were retained for further analysis (Fig. 6).

Fig. 6
figure 6

Community locations in the province of Alberta, Canada. Black markers denote communities selected for analysis (n = 913) and green markers denote omitted communities (n = 590). For reference, Alberta’s Forest Protection Area (FPA) is shown with shading along with the administrative boundaries of the province’s ten Forest Areas

Full size image

In total, we evaluated 986,040 directional segments. Of the 913 communities analyzed, more than half (n = 519, 56.8%) had no viable directional trajectory segments. Of the 394 remaining communities, 136 had at least one continuous viable trajectory spanning the full 0–15-km distance from the community centroid to the perimeter of the circular assessment area, and 211 communities had at least one continuous viable trajectory spanning 5–15 km from the community centroid. Moderately-sized communities (i.e., locations with a population > 1000 during the fire season) with at least ten viable directional trajectories 5–15 km from centroid are listed in Table 2.

Table 2 Count of viable directional trajectories 5–15 km from centroid, by community. Only larger communities are shown (i.e., towns and villages) as well as selected hamlets and localities with populations > 1000
Full size table

The community with the highest number of continuous viable directional trajectories spanning the 5–15-km distance was Nordegg, a hamlet in west-central Alberta where 59% (n = 213) of the 360 possible wildfire directional trajectories into the community were viable at the 5–15-km distance from community centroid (Fig. 7A). Nordegg also had 154 continuous directional trajectories spanning the full 0–15-km distance from centroid (Fig. 7B). The Town of Hinton had viable continuous wildfire pathways for 147 different directional trajectories at the 5–15-km distance from community centroid and significant potential for negative wildfire impacts, given the population exceeds 10,000. The Town of Jasper had 136 viable directional trajectories spanning 5–15 km from centroid (Fig. 8). Viable fire pathways into Jasper were oriented predominantly from the west-northwest (Fig. 8).

Fig. 7
figure 7

A Community of Nordegg directional trajectories that exhibited alignment with high-exposure areas (i.e., 80% of the trajectory length intersected high exposure), by 5-km segment. B Directional trajectories that exhibited high-exposure alignment continuously for all three distance segments (0–5 km, 5–10 km, and 10–15 km)

Full size image
Fig. 8
figure 8

A Community of Jasper directional trajectories that exhibited alignment with high-exposure areas (i.e., less than 1/5th of the trajectory length intersected low exposure), by 5-km segment. B Directional trajectories that exhibited high-exposure alignment continuously for all three distance segments (0–5 km, 5–10 km, and 10–15 km)

Full size image

The frequency or load of viable directional pathways for all 913 communities analyzed is summarized by operational Forest Areas in Fig. 9. The y-axis indicates the frequency of communities within a given Forest Area management unit that has a continuous viable trajectory at the 5–15-km distance from centroid in a given direction. We can see that directionality is highly variable. For example, the Slave Lake Forest Area has multiple communities with potential fire pathways aligned with northeast winds, whereas the adjacent Lac La Biche Forest Areas exhibit no viable pathways of fire encroachment from that direction. There is also variability in the magnitude of the load across the Forest Areas, with Forest Areas in the northern half of the province exhibiting relatively light loads of viable directional pathways compared with Edson and Rocky Mountain House Forest Areas.

Fig. 9
figure 9

Directional vulnerabilities summarized by Forest Area administrative units. Radial graphs show the number of communities where a given directional trajectory is aligned with high exposure, in this case limited to the 5–15-km segment of the trajectory. Yellow shading shows the count of communities in the Forest Area where there is alignment with high exposure at the 5–15-km segment of the trajectory. Frequencies (i.e., y-axis values) range from 0 to 9. Circular gridlines are displayed in intervals of 2

Full size image

4 Discussion

We used a simple metric of landscape fire exposure (i.e., Beverly et al. 2021) to develop a systematic and standardized approach for assessing directional vulnerability to wildfire within a circular assessment area centered on locale of interest. When applied to 913 small- to moderately-sized communities in the province of Alberta, Canada, we found that communities exhibited highly unique and varied patterns of directional vulnerabilities to wildfire encroachment.

The amount and distribution of viable wildfire trajectories surrounding a given community is relevant to both fire response decisions and local mitigation planning. For example, assessments of directional vulnerabilities could be used to prioritize newly detected fires for response action and resource allocation. By overlaying a newly reported fire coordinate and observed wind direction on our directional assessments, a fire manager can easily and immediately identify alignment of the fire with potential direct pathways into a community and prioritize that fire for dispatched resources accordingly. Fire growth modeling simulations (e.g., Tymstra et al. 2010) can also be used to assess the potential spread of a newly reported fire; however, the time and expertise required to conduct and interpret fire growth simulations can be prohibitive. Fire management agencies dealing with high fire loads across large landscape areas may not have resources to calibrate the model for each local fire environment and conduct sensitivity analysis to ensure predictions are realistic.

Likewise, in situations where resources are limited and multiple fires are ignited under extreme fire weather conditions, and notably extreme winds, the timeline for choosing which fires will receive limited suppression resources is extremely narrow, and modeling exercises in these circumstances can be impractical and potentially unreliable. Fire growth simulations utilize empirical fire behavior prediction models (i.e., Forestry Canada Fire Danger Group 1992) that are based on many simplifying assumptions, as well as a large number of inputs that make them prone to accumulated uncertainties. In contrast, our simple descriptive directional vulnerability assessments are based on a single input (i.e., fuels) and can be produced for all communities in a given jurisdiction using rudimentary GIS analysis conducted well in advance of the fire season. Our directional assessments are based on simplifying assumptions, but these assumptions have been checked against empirical data (i.e., real wildfires) reported herein and in Beverly et al. (2021). When combined with current and forecast weather conditions, our directional vulnerability assessments can provide an instantaneous red flag warning for informing suppression priorities and related response actions, such as evacuation decisions.

Directional assessments could also be used to inform proactive strategic planning activities conducted over longer decision horizons, such as planning fuel reduction treatments around a community (e.g., Beverly et al. 2020), proactively identifying candidate locations for supporting operational tactics such as fireline construction in the event of a fire (e.g., O’Connor et al. 2016), or identifying evacuation vulnerabilities associated with the transportation network (e.g., Dye et al. 2021). To explore the interaction between biophysical wildfire directional vulnerability and evacuation vulnerability, our directional assessments were used as an input to transportation network analysis to identify transportation vulnerabilities for strategic evacuation planning in a recent study (i.e., Al Zahid 2022).

The distribution of viable directional trajectories around a community also provides important insight into the scope of the activities needed to remove or reduce directional vulnerabilities. For example, a community with a small number of viable directional trajectories grouped together and spanning an overall limited directional extent represents a fundamentally different strategic planning challenge than one with abundant directional vulnerabilities distributed evenly around the community. Indeed, it could be argued that in the latter case, responding quickly to fires and preparing to safely evacuate residents in the event of a fire may be the most prudent course of action, due to the vast scope of fuel reduction treatments that would be required to fully disrupt potential fire pathways.

Beyond the assessment of individual communities, we have illustrated how directional vulnerabilities can be compiled for large administrative areas. Our results showed the pattern of directional vulnerabilities across Forest Area administrative units in Alberta varied from region to region. These strategic regional assessments of directional vulnerabilities within a given Forest Area (Fig. 9) can be combined with forecast wind direction to alert fire management agencies of potential vulnerabilities across their management areas and aid in day-to-day resource allocation decisions. For example, if forecast winds are predominantly southwesterly in the Calgary Forest Area, there is limited potential for fire encroachment on communities within that Forest Area (Fig. 9). Whereas, if winds are forecast to be predominantly northwesterly in the Edson Forest Area, there is vulnerability of fire encroachment for communities in that region.

Our regional directional vulnerability summaries highlight the challenges of predicting when and where a fire might encroach on a given community. For a fire disaster to unfold, an ignition must occur not only upwind, but also positioned along a viable fire pathway into a community that is aligned with the concurrent wind direction. Ignition locations are notoriously difficult to predict, and wind directions are by far the most highly variable aspect of the fire environment. Historical patterns of ignitions can be mapped (e.g., Coogan et al. 2022), and predominant wind directions around a location can be summarized from historical weather data (e.g., Huffman et al. 2021); however, the extent to which historical frequencies of ignitions and wind directions are relevant to assessing current vulnerability is questionable, especially under climate change.

When conditions are uncertain, likelihoods are considered ill-suited for characterizing the risk of rare events, and examination of possible outcomes is recommended as an alternative (Stirling 2007; Flage et al. 2014). Areas assessed as having high-exposure highlight where wildfire transmission is possible, irrespective of stochastic fire controls (i.e., ignition or weather). Likewise, our directional trajectories identify possible pathways of wildfire encroachment into a community. These possible fire outcomes can be used to develop what-if scenarios for futures and foresight analysis, a process that is increasingly being used to inform strategic risk management (Riddell et al. 2020). In scenario planning, the objective is to explore possible futures rather than attempting to predict what will happen (Shoemaker 1995).

Our directional vulnerability assessments focus on a single input that we can know with relative certainty in advance of a fire: fuels. While this eliminates the inherent uncertainties imbedded in more complex and computationally intensive fire growth models, the approach is not without limitations. Firstly, the landscape fire exposure metric that we used to define viable fire trajectories around communities  was generated from relatively coarse-scale fuel data (i.e., 100 m × 100 m grid cells). Ideally, this landscape-scale assessment is used to prioritize communities for a paired, finer-resolution exposure assessment in the localized area around the community (i.e., Beverly et al. 2010), which would permit a more sophisticated and detailed assessment of viable fire trajectories immediately adjacent to the built environment. We are currently working to assess a number of case study communities in Alberta at that finer scale, including Nordegg and Hinton.

Second, we know that under anomalous conditions, like extreme drought, fires could potentially burn areas assessed as having low exposure and travel along trajectories that we did not consider viable. Beverly et al. (2021) reported that some low-exposure locations in Alberta subsequently burned, but this was generally limited to a small proportion of the total observed burned area. It is not surprising that low-exposure areas were sometimes observed to burn, given that exposure was calculated for coarse 1-ha pixels whereas burned areas present with comparatively detailed and irregular morphologies. Fire spread through low-exposure areas is also expected when the underlying land cover data used to calculate exposure is inaccurate or when assumptions about hazardous fuels are inconsistent with local conditions. While it is possible for areas assessed as having low exposure to burn, we know that areas burned by wildfire in Alberta occur preferentially in high-exposure areas, and this was confirmed by our sample trajectories delineated within historical fire perimeters, which suggests that the viable fire trajectories we have mapped are reliable for informing fire management decisions.

In this study, we assessed directional wildfire vulnerability for community locations; however, the approach could also be applied to any mapped value or location of concern such as an industrial facility, infrastructure, critical habitat, or sensitive watershed. Our future research will explore the application of our directional assessments of wildfire vulnerability to these other types of point features and will explore methods to describe the statistical properties of the directional data (e.g., Huffman et al. 2021). Use of point locations limited our analysis to community sizes that were conducive to representation with a centroid. For larger communities, the same methods could be used to assess vulnerabilities of multiple individual neighborhoods using a series of assessment points. Our future work will explore a custom approach for distributing potential fire trajectories in the landscape surrounding polygons rather than point locations.

Finally, the novel spatially-explicit rosette diagrams that we developed to assess directional patterns of high-exposure areas in this study could be applied to any land-based environmental data to describe the circular distribution of presence/absence data within distance bands encircling a feature of concern. For example, while not the focus of our work, spatially-explicit rosettes could be used to characterize the directional distribution of high-suitability habitat around a planned industrial facility to guide site planning. With respect to wildfire, our directional assessments are not limited to exposure as an input and could be applied to any map of spatial variation in fire potential generated by any means, provided a suitable presence/absence criterion for viable trajectories could be established.