Short-term impacts of operational fuel treatments on modelled fire behaviour and effects in seasonally dry forests of British Columbia, Canada

Study sites

We partnered with five community forests in the Kootenay–Boundary Natural Resource Region of BC: Creston, Harrop–Procter, Kaslo and District, Nakusp and Area Community Forests, and Slocan Integral Forestry Cooperative (Fig. 1). These forest tenures range from ~9000 to 33,000 ha, each forming part of the wildland–urban interface of a rural community. These community forests primarily occur in the low-to-mid elevation (i.e. 400–1500 m above sea level) Interior Cedar–Hemlock (ICH) biogeoclimatic ecosystem classification zone, which is characterised by snowy winters and hot dry summers (Ketcheson et al. 1991; Braumandle et al. 1992). Mean temperatures are below 0°C for the coolest winter months and above 15°C for the warmest summer months. Mean annual precipitation ranges from 500 to 1200 mm, with 25–50% as snow (Ketcheson et al. 1991). The forests in the ICH zone have diverse tree species, including western red cedar (Thuja plicata Donn ex D. Don), western hemlock (Tsuga heterophylla (Raf.) Sarg.), western larch (Larix occidentalis Nutt.), interior Douglas-fir (Pseudotsuga menziesii (Mirb.) Franco var. glauca (Beissn.) Franco), ponderosa pine (Pinus ponderosa Douglas ex Lawson), lodgepole pine (Pinus contorta Douglas ex Loudon var. latifolia Engelm.) and paper birch (Betula papyrifera Marshall) (Ketcheson et al. 1991). This seasonally dry mixed-conifer forest type historically experienced mixed-severity fire regimes that have been disrupted by fire exclusion for the past century (Nesbitt 2010; Marcoux et al. 2013, 2015; Greene and Daniels 2017; Chavardès et al. 2022). Contemporary stands in the community forests are dense and continuous. For all five forests, the provincial wildfire threat ratings are high to extreme (British Columbia Wildfire Service 2021), making them a priority for proactive fuel mitigation treatments (British Columbia Community Forest Association 2024).

Fig. 1.

Study site locations throughout the Kootenay–Boundary Natural Resource Region of British Columbia, Canada. The five community forests are Creston Community Forest (CCF), Harrop–Procter Community Forest (HPCF), Kaslo and District Community Forest (KDCF), Nakusp and Area Community Forest (NACFOR), and Slocan Integral Forestry Cooperative (SIFCo). Basemap attribution: map tiles by Stamen Design, under Creative Commons (CC) BY 3.0 (available at https://creativecommons.org/licenses/by/3.0/ [accessed 29 December 2021]). Data by OpenStreetMap, under ODbL (available at https://www.openstreetmap.org/copyright [accessed 29 December 2021]). Natural Resource Region boundaries attribution: contains information licensed under the Open Government License – British Columbia (available at https://www2.gov.bc.ca/gov/content/data/open-data/open-government-licence-bc [accessed 27 December 2021]). Road attribution: contains information licensed under the Open Government License – Canada (available at https://open.canada.ca/en/open-government-licence-canada [accessed 28 December 2021]). Provincial boundaries attribution: Boundary Files, 2011 Census. Statistics Canada Catalogue no. 92-160-X.

Development of fuel treatment prescriptions in BC

In BC, the guidelines and criteria for fuel treatments set by the BC Wildfire Service are rapidly evolving. When the fuel treatment prescriptions were developed in 2020 and 2021, the community forest managers primarily relied on broad guidance documents from the BC Wildfire Service (British Columbia Wildfire Service 2018, 2020). These guidance documents outlined common principles for fuel reduction treatments, such as reduce surface fuel loads and limit the vertical and horizontal continuity of crown fuels. The BC Wildfire Service provided an optional online worksheet for managers to calculate wildfire intensity (Byram 1959) and critical surface fire intensity for initial crown combustion (Van Wagner 1977; British Columbia Wildfire Service 2024). Outside this provided worksheet, the community forests we collaborated with did not utilise any stand-level fire behaviour models in the development of their fuel treatment prescriptions.

Treatment units and fuel treatment types

Eleven treatment units across the five community forests were used in this study (Table 1). Some treatment units included multiple stands based on tree species composition and planned fuel treatment type, for a total of 26 stands. The fuel treatment prescriptions for all stands were developed by community forest managers using guidance documents provided by BC Wildfire Service (British Columbia Wildfire Service 2018, 2020) and implemented between fall (autumn) 2021 and spring 2022.

We categorised the fuel treatments implemented on the community forests into five different types based on thinning, pruning and residue fuel management (Table 2). Thinning included two levels of thinning from below: low or high intensity (Fig. 2). The post-treatment target tree density varied depending on the starting stand condition and the specific management prescription, but was generally 400–1400 stems ha⁻¹ of well-spaced trees for low-intensity thinning versus 50–350 stems ha⁻¹ for high-intensity thinning. Operational implementation of these thinning levels was achieved via operator execution of guidelines stated in treatment prescriptions rather than via pre-marking of trees for removal or retention. In these prescriptions, trees of any size or species were permitted to be removed for the safety of the workers or the efficiency of the operations. Live and healthy deciduous trees of all sizes were generally retained, as deciduous trees tend to be less flammable than coniferous trees (Pausas et al. 2017), and did not contribute to the target densities. Among coniferous trees, mature, fire-resistant interior Douglas-fir, western larch and ponderosa pine were preferentially retained. For the low-intensity thinning prescriptions, trees were removed by hand falling until a target retained density was achieved. Generally, only non-merchantable coniferous trees with diameter outside bark at breast height (1.3 m above ground; DBH) <17.5 cm considered not suitable for sawn wood products in BC were removed. For the high-intensity thinning prescriptions, tree removals via harvesting machines began with non-merchantable trees and then extended to merchantable trees to reach a target density.

Fig. 2.

Examples of low- and high-intensity thinning from below. Low-intensity thinning example (top row) is from Harrop–Procter Community Forest and high-intensity thinning example (bottom row) is from Slocan Integral Forestry Cooperative. Orange stars represent anchor points for the paired pre- (left column) and post-treatment (right column) photos. All photos in the figure were taken by Kea Rutherford.

For prescriptions that included pruning, low branches of retained coniferous trees were removed to achieve a live crown base height of 2.5 or 3 m. To avoid compromising small coniferous trees, 40% live crown was retained for these trees rather than pruning to a specific height above ground.

Lastly, residue fuel management included two categories: pile burning or chipping. In the stands where residue fuel was chipped, the wood chips were evenly distributed across the stand to a target depth of 5–10 cm.

Field measurements

We established 43 sampling points across the 26 stands (Table 1). Where community forest personnel had previously established a systematic grid of sampling points, a few of these were randomly selected for measurement. In the other stands, sampling points were selected via simple random sampling.

Field measures were taken before and after treatment in the summers of 2021 and 2022, respectively. At each sampling point, we measured slope, aspect, elevation and key components of the fuelbed. Canopy trees with a DBH ≥12.5 cm were measured within a circular fixed-area plot (Gillis et al. 2005). If tree density was <1000 stems ha⁻¹, we used a 0.04-ha (11.28 m radius) plot; if tree density was ≥1000 stems ha⁻¹, we used a 0.02-ha (7.98 m radius) plot. Similarly, subcanopy trees with a DBH ≥5 cm and <12.5 cm were measured within a smaller subplot. If tree density was <4000 stems ha⁻¹, we used a 0.01-ha (5.64 m radius) plot; if tree density was ≥4000 stems ha⁻¹, we used a 0.005-ha (3.99 m radius) plot. For all standing trees, the species, DBH, total height, live crown base height, dead crown base height, crown position (i.e. progressing downward from dominant trees in the upper canopy through codominant and intermediate trees to suppressed trees in the lower canopy) and status (i.e. live, dead with decay class; USDA Forest Service 2019) were recorded. Surface fuels were measured using the line-intercept method (Brown 1974) along two perpendicular 30-m transects crossed at the mid-point. We tallied 1-h surface fuels (i.e. diameter ≤0.635 cm). For all 10-, 100- and 1000-h surface fuels (i.e. diameter >0.635 cm), we measured the diameter at the point of intersection and recorded decay class and, when possible, species. Surface fuel loads were calculated using published equations (Brown 1974; Beaufait et al. 1975; Ryan and Pickford 1978; US Forest Products Laboratory 1980; Woodall and Monleon 2008). For all 1-h surface fuels and for 10-, 100- and 1000-h surface fuels of unknown species, we weighted the coefficients required to calculate loadings by percentage basal area by species (Stephens 2001). Canopy cover was measured along the two line transects using a tube densitometer (Jennings et al. 1999). We measured the distance along each transect where shrubs intersected the line, which we used to calculate percentage shrub cover (Canfield 1941). A 0.01-ha circular subplot was established at the end of one of the line transects, in which regenerating trees (DBH <5 cm and height >1.3 m) were counted by species and status (i.e. live, dead) and percentage covers of grass and moss were ocularly estimated. The same field measurements were taken at the same sampling points pre- and post-treatment, except the DBH and total height of each retained tree, which we assumed did not change pre- to post-treatment (i.e. only 1 year of growth).

Stand attributes

The fuel treatments prescribed by the community forest managers aimed to reduce or rearrange surface, ladder and crown fuels. For each sampling point, key stand attributes were compiled from the pre- and post-treatment field data (Table 3). Surface fuels were described by fine fuel load. Ladder fuels were described by the density of small live trees, average live crown base height and average dead crown base height. Crown fuels were described by the density of medium plus large live trees and the relative composition of fire-resistant species.

Modelling stand-level fire behaviour and effects

To assess potential stand-level fire behaviour and effects before and after treatment, we used the BC variant of FVS-FFE (Rebain et al. 2010; ESSA Technologies Ltd. 2013). The FFE for the BC variant of FVS uses the FFE originally developed for the Inland Empire variant, which has a recommended geographic range of northern Idaho, western Montana and eastern Washington, an area directly south of the US–Canada border with very similar forest types to our study sites (Ketcheson et al. 1991). To model potential fire behaviour, FVS-FFE links Rothermel (1972, 1991) surface and crown fire rate of spread models with Van Wagner (1977, 1993) crown fire transition and propagation models (Supplementary Material A: Table S1 summarises component models in FVS-FFE). A limitation of this modelling approach is that it generally underpredicts fire behaviour and effects (Fulé et al. 2001; Hall and Burke 2006; Cruz and Alexander 2010), with underprediction more pronounced for the pre-treatment stands (Foster et al. 2020; Stephens et al. 2023). Nevertheless, it has proved to be an effective approach for evaluating relative differences among fuel treatments in seasonally dry forests of the western US (e.g. Low et al. 2023; Stephens et al. 2023; Radcliffe et al. 2024). A strength of FVS-FFE is that it uses 53 standardised surface fuel models that include a range of fire-carrying fuel types (i.e. grass, grass–shrub, shrub, timber litter, timber–understorey and slash; Anderson 1982, Scott and Burgan 2005). These surface fuel models can represent important changes in the type, loading and compactness of surface fuels resulting from treatments. Therefore, we determined that FVS-FFE was a suitable stand-level model to evaluate the relative impacts of different types of fuel treatments on potential fire behaviour and effects in our study system.

Inputs to FVS-FFE included a tree list (i.e. species, status, DBH, total height, live crown ratio and stems per hectare for each tree), slope, aspect, elevation and weather conditions for each plot. For weather conditions, we matched the closest fire weather station within a 300-m elevation range to each community forest. Using archived weather data from 2007 to 2021 (15 years; British Columbia Wildfire Service 2022), we calculated the 90th percentile of temperature to be consistent with standards for fuel treatments set by the BC Wildfire Service (British Columbia Wildfire Service 2023b). We averaged the wind speeds concurrent with the 90th percentile temperature (Table 4). We set fuel moisture to the pre-defined, variant-specific ‘very dry’ group in FVS-FFE (out of four groups: very dry, dry, moist and wet), which best aligned with high fire risk conditions in our study area.

From the 53 standardised surface fuel models built into FVS-FFE, pre-treatment surface fuel models were assigned based on measured shrub, grass and moss cover, plus fine surface fuel loads (Table 5; Vaillant et al. 2009). Post-treatment fuel models depended on treatment type. For high-intensity thinning combined with pile burning (fuel treatment Types 1 and 2), timber litter and slash models were assigned to represent disruption of the understorey and measured generation of residue fuel. For low-intensity thinning combined with pile burning (fuel treatment Types 3 and 4), the post-treatment surface fuel conditions were the same as the pre-treatment conditions. For low-intensity thinning combined with chipping (fuel treatment Type 5), timber litter models were assigned to emulate the compact structure of chipped fuelbeds. To account for inherent uncertainty in assigning discrete surface fuel models, we assigned a ‘low’ and a ‘high’ surface fuel model to each plot, both pre- and post-treatment. The low surface fuel model has a lower relative spread rate and flame length than the paired high surface fuel model (Chiono et al. 2017; Low et al. 2023). The low and high surface fuel models were given equal weights in the FVS-FFE simulations to yield one pre- and one post-treatment outcome per plot.

Outputs of FVS-FFE include modelled fire behaviour (i.e. probability of torching (P-Torch) and crowning index) and fire effects (i.e. tree mortality) for each plot pre- and post-treatment (Table 3). P-Torch has been shown to reflect changes in the potential of passive crown fire more realistically than the alternative torching index metric because it does not require a calculation of stand-level canopy base height, which can be difficult to accurately represent (Rebain et al. 2010; Stephens et al. 2023). FVS-FFE is a powerful tool for making relative comparisons among fuel treatments (e.g. Low et al. 2023; Stephens et al. 2023); however, absolute values of the FVS-FFE outputs must be interpreted with caution. In particular, values of the crowning index can exceed realistic maximum wind speeds, indicating that the stand is highly resistant to active crown fire (Stephens et al. 2009). Based on historical weather data from the Kootenay region (Government of Canada 2022), we selected a threshold wind speed of 60 km h⁻¹. FVS-FFE outputs above this threshold were reassigned a value of 60 km h⁻¹.

Meta-models of fire behaviour and effects metrics

To examine the impacts of the five fuel treatment types on P-Torch, crowning index and tree mortality metrics derived from FVS-FFE, we fitted linear mixed-effects meta-models (Pinheiro and Bates 2000) using the ‘nlme’ package (Pinheiro et al. 2023) in R (R Core Team 2023). This is a meta-modelling approach because FVS-FFE outputs were used as the response variables (e.g. Fulé et al. 2012; Ahmed et al. 2020). For all three models, the response variable was the change in the fire metric from pre-treatment to post-treatment. Fuel treatment type was included as an explanatory variable in all models. Other explanatory variables included represented starting stand conditions calculated using pre-treatment field measurements, namely: average tree height for P-Torch, density of live trees for crowning index and Curtis Relative Density (RD; diameter and stems per hectare stand density metric; Curtis 1982) for tree mortality. Given the modest number of measured plots resulting in 43 records for fitting the linear mixed-effects models (Table 2), only one starting stand condition variable was chosen for each response variable, selected based on our knowledge of the fire metric. A two-way interaction between the starting stand condition variable (i.e. continuous variable) and fuel treatment type (i.e. class variable) was tested for inclusion in each model. Random intercepts for treatment unit and forest were included in the models. These random treatment and forest effects account for possible spatial correlation among plots from the same treatment unit and community forest.

For each response variable, we selected the best random-effects structure by performing likelihood ratio tests (α = 0.05) using the restricted likelihood and the ‘lmtest’ R package version 0.9.40 (Zeileis and Hothorn 2002). Proceeding with the selected random-effects structure, we fitted a model with fuel treatment type, the starting stand condition variable and the possible interaction. We selected the best fixed-effects structure using manual backward elimination (Kutner et al. 2005), starting with testing whether the interaction term could be dropped. To decide whether a variable was dropped from the model, we performed a likelihood ratio test (α = 0.05) using the full likelihood fit of the model. For each model, we checked that the assumptions of linear models were met by examining residual, normality and fitted line plots.

For models where the interaction with fuel treatment type was not significant, multiple comparisons among treatment means adjusted to the average of the starting stand condition variable were obtained using the ‘emmeans’ package in R (Lenth 2023) for: (1) high- versus low-intensity thinning treatments; (2) presence or absence of pruning within high-intensity thinning treatments; (3) presence or absence of pruning within low-intensity thinning treatments; and (4) pile burning and chipping residue fuel within low-intensity thinning treatments. Significant differences were determined by an α-level of 0.05, adjusted for multiple comparisons using the common Bonferroni P-value adjustment method. Only comparisons relevant for the specific fire metric were evaluated.

Summary statistics of stand attributes within fuel treatment types

As expected, the densities of large-diameter trees substantially decreased following the high-intensity thinning treatments (i.e. fuel treatment Types 1 and 2; Fig. 3a). After treatment, the densities of large trees ranged from 0 to 325 stems ha^–1, depending on the pre-treatment stand condition and the specific management prescription. For the low-intensity thinning treatments (i.e. fuel treatment Types 3, 4, and 5), the densities of large trees remained mostly unchanged, as prescribed. The densities of medium- and small-sized trees substantially decreased following treatment for all five treatment types (Fig. 3b, c). For the low-intensity thinning treatments, the density of medium-sized trees ranged from 0 to 700 stems ha^–1 post treatment. Variation among sampling points reflected prescriptions that stipulated removals of trees below 12.5, 15 or 17.5 cm DBH, depending on pre-treatment stand conditions.

Fig. 3.

Measured stand densities stratified by tree size (rows) and fuel treatment type (columns). Tree sizes are (a) large (DBH ≥17.5 cm); (b) medium (DBH 7.5–17.5 cm); and (c) small (DBH <7.5 cm and height >1.3 m). Blue represents pre-treatment and green represents post-treatment values. Within box plots, lines represent median values and stars represent mean values. For each treatment type, n is the number of sampling points.

Changes in the relative composition of fire-resistant species reflected pre-treatment stand composition and the type of fuel treatment applied (Fig. 4a). For fuel treatment Type 1 (i.e. high-intensity thinning treatment), operators had a greater opportunity to preferentially select fire-resistant species for retention. Thus, the relative composition of fire-resistant species generally increased following treatment, with much higher variability pre- versus post-treatment. Stands that received fuel treatment Types 2 and 3 had low proportions of fire-resistant species before treatment, providing minimal opportunity to prioritise the retention of fire-resistant species so that the relative composition remained low after treatment. Stands that received fuel treatment Types 4 and 5 (i.e. low-intensity thinning treatments) had high variability in the relative composition of fire-resistant species before and after treatment. For these treatments, trees were selected for removal primarily based on size limits rather than species.

Fig. 4.

Measured stand attributes (panels a–d), summarised by fuel treatment type (columns within each panel). Blue represents pre-treatment and green represents post-treatment values. Within box plots, lines represent median values and stars represent mean values. For each treatment type, n is the number of sampling points.

Average live crown base height was highly variable before treatment in all fuel treatment types (Fig. 4b). In stands that received high-intensity thinning treatments, all sampling points had a starting average live crown base height >6 m, well above the 2.5 or 3 m pruning target. Similarly, in stands that received low-intensity thinning treatments, most sampling points had a starting average live crown base height >3 m. Regardless of the presence or absence of pruning, average live crown base height generally increased following treatment owing to the retention of larger trees, which tend to have higher live crown base heights.

Pruning also removed dead branches below the 2.5 or 3 m pruning target. Stands that received high-intensity thinning treatments had high variability in average dead crown base height before treatment (Fig. 4c). For fuel treatment Type 1, variability was greater after treatment. For fuel treatment Type 2, average dead crown base height generally increased but variability was similar after treatment. In stands that received low-intensity thinning treatments, average dead crown base height generally increased following removal of most small trees, but variability remained moderate before and after treatment.

Fine fuel load substantially increased with the high-intensity thinning treatments that removed a greater number of trees using harvesting machines, which generated a large amount of residue fuel as small trees, treetops and branches were scattered on the forest floor (Fig. 4d). The increase of fine residue fuel was reduced through pile burning with varying degrees of success, as demonstrated by the high variability in fine fuel load after treatment, especially within fuel treatment Type 1. For low-intensity thinning treatments combined with pile burning (i.e. fuel treatment Types 3 and 4), the ranges in fine fuel load were similar pre- and post-treatment owing to the removal of residue fuel through pile burning. Fine fuel load generally increased with the low-intensity thinning treatment combined with chipping (i.e. fuel treatment Type 5), as expected.

Differences in modelled fire behaviour and effects among fuel treatment types

For P-Torch, the final meta-model had fuel treatment type and average tree height as fixed effects (Supplementary Material B: Tables S1 and S2). The final meta-model for crowning index had a random treatment unit effect as well as fuel treatment type and density of live trees as fixed effects (Supplementary Material B: Tables S3 and S4). The final meta-model for tree mortality had fuel treatment type and Curtis RD as fixed effects (Supplementary Material B: Tables S5 and S6). The interaction between treatment and the specific starting stand condition variable was not significant in any of the meta-models. Fit statistics for all final meta-models are provided in Supplementary Material C.

P-Torch generally decreased following treatment across all fuel treatment types, indicating lower relative potential for passive crown fire (Fig. 5a). Using the meta-model, the change in P-Torch from pre- to post-treatment was not significantly different between low- and high-intensity thinning treatments (Table 6). Within high-intensity thinning treatments, there was no significant difference in the change of P-Torch between pruning and not pruning. Within low-intensity thinning treatments, neither pruning nor residual fuel management approach had a significant effect on the change in P-Torch.

Fig. 5.

Fire behaviour and fire effects metrics output from FVS-FFE (panels a–c), summarised by fuel treatment type (columns within each panel). Blue represents pre-treatment and green represents post-treatment values. Within box plots, lines represent median values and stars represent mean values. For each treatment type, n is the number of sampling points.

The crowning index generally increased with treatment across all fuel treatment types, representing lower potential of active crown fire (Fig. 5b). The change in the crowning index from pre- to post-treatment was significantly different between low- and high-intensity thinning treatments. On average, the crowning index increased by 12 km h⁻¹ for high-intensity relative to low-intensity thinning (Table 6).

Modelled tree mortality generally decreased following treatment across all fuel treatment types (Fig. 5c). The change in modelled tree mortality from pre- to post-treatment was significantly different between low- and high-intensity thinning treatments. On average, modelled tree mortality decreased by 12% for high-intensity relative to low-intensity thinning (Table 6). Within low-intensity thinning treatments, there was no significant difference in the change of tree mortality between chipping and pile burning.

Modelling stand-level fire behaviour and effects in seasonally dry montane forests

Modelling stand-level fire behaviour and effects proved challenging owing to inherent limitations of existing models. Research conducted in montane, mixed-conifer forests of the western US commonly employs FVS-FFE, which links Rothermel (1972, 1991) surface and crown fire rate of spread models with Van Wagner (1977, 1993) crown fire transition and propagation models. A limitation of this modelling approach is that it generally underpredicts fire behaviour and effects, which Cruz and Alexander (2010) attribute, in part, to inherent underprediction in the Rothermel (1972, 1991) rate of spread models and incompatible linkages between Rothermel’s (1972) surface fire and Van Wagner’s (1977) crown fire initiation equations.

Given these limitations, we considered using the Crown Fire Initiation and Spread (CFIS) system developed by Alexander et al. (2006a, 2006b). However, we identified incongruencies that limited its application for assessing treated montane forests in the ICH zone of southeastern BC. Specifically, CFIS does not include a built-in surface fire rate of spread model, which prevents the assessment of surface fire effects (e.g. mortality). Additionally, CFIS represents surface fuels as three categorical consumption classes (<1, 1–2 and >2 kg m^–2; Alexander et al. 2006b). For many forests in BC, the Canadian Forest Fire Behaviour Prediction System (FBP) can be used to estimate surface fuel consumption and select an appropriate class (Beverly et al. 2020; Perrakis et al. 2023). However, the FBP fuel types poorly represent the ICH forests in our study area (Parisien et al. 2013), particularly managed stands in the seasonally dry montane forests of interior BC (Baron et al. 2024). In its current form, CFIS does not represent the nuanced changes in type, loading and compactness when surface fuels are managed through reduction (e.g. pile burning, broadcast burning) and rearrangement (e.g. chipping).

Ultimately, we determined that FVS-FFE was the best currently available stand-level model to evaluate the relative impacts of alternative types of fuel treatments on potential fire behaviour and effects in our study system. We moderated underprediction bias by FVS-FFE to the extent possible through careful surface fuel model selection based on detailed quantitative data on vegetation cover and fine surface fuel loads, an approach that substantially improves fire behaviour predictions relative to those derived using the default surface fuel model selection in FVS-FFE (Collins et al. 2013). We recognise that fire behaviour and effects metrics derived from FVS-FFE are not precise predictions. To avoid overinterpretations, we summarised our field-based measurements of stand attributes and coupled them with meta-models of fire behaviour and effects metrics derived from FVS-FFE to provide insights into the potential relative impacts of the five alternative fuel treatments implemented by community forest managers in our study sites. Below, we discuss key interpretations from the meta-modelling results and identify priorities for future management and research.

Thinning treatments may reduce potential of crown fire and tree mortality

Low- and high-intensity thinning from below are common fuel treatments in the community forests of BC (British Columbia Community Forest Association 2024). In this study, both levels of thinning intensities generally decreased modelled potential of passive crown fire by mitigating dense understories of shade-tolerant western red cedar, western hemlock and interior Douglas-fir through the removal of small- and medium-sized trees. The removal of understorey trees reduces ladder fuels, hindering vertical fire propagation from the surface into the crown (Agee and Skinner 2005; Harrod et al. 2009; Stephens et al. 2009). The high-intensity thinning treatments reduced the potential for active crown fire more than the low-intensity thinning treatments. Removal of some larger trees by high-intensity thinning also disrupts horizontal continuity of crown fuels, making the stand less susceptible to tree-to-tree crown fire spread (Agee and Skinner 2005; Johnson et al. 2011).

The low- and high-intensity thinning treatments were designed to retain larger trees to increase resistance to fire by lowering tree mortality in the advent of wildfire. Modelling outcomes indicated that both types of thinning generally decreased potential tree mortality; however, the high-intensity thinning treatments led to lower potential tree mortality. The high-intensity thinning involved removal of most of the smaller trees before extending into larger trees, favouring the retention of mostly the largest trees in the stand. Larger trees tend to be more fire-resistant, because they are taller and less likely to endure substantial crown scorch (Peterson and Ryan 1986; Agee and Skinner 2005) and have thicker bark that makes them less susceptible to cambial girdling (Ryan and Reinhardt 1988; van Mantgem and Schwartz 2003). Additionally, some of the thinning prescriptions were designed to prioritise the retention of fire-resistant species to increase stand resistance to fire. Interior Douglas-fir, ponderosa pine and western larch tend to have thicker bark and higher crowns than other common conifer species in the ICH zone, making them particularly fire-resistant (Agee 1993; Jain et al. 2012). The substantial increase in relative composition of fire-resistant species for fuel treatment Type 1 (i.e. high-intensity thinning) demonstrates the potential of more intensive thinning to favourably shift the composition of species.

In this study, the community forest managers mostly prescribed low-intensity thinning treatments (20 out of 26 stands), partially in response to public opposition to high-intensity thinning treatments (Copes-Gerbitz et al. 2020; Prichard et al. 2021). However, the modelling results of the present study indicate that high-intensity thinning in the dense stands of the ICH zone represented in this study may reduce passive and active crown fire activity and resultant tree mortality, even during dangerous fire weather conditions (e.g. 90th percentile). When developing fuel treatments, managers must consider how to prevent the initiation of crown fire within a stand, as well as how to reduce the behaviour of an already initiated crown fire spreading from adjacent stands (Scott and Reinhardt 2001). In BC, a specific goal of treatments in the wildland–urban interface is to moderate fire behaviour so that crown fires transition to surface fires, providing greater opportunity for safe and successful suppression tactics (British Columbia Wildfire Service 2020; British Columbia Community Forest Association 2024). The modelling outcomes show that stands that undergo low-intensity thinning may be fairly resistant to the initiation of crown fire, but may not necessarily aid the cessation of crown fire. In contrast, stands that receive high-intensity thinning may be more resistant to the initiation and spread of crown fire. Given that there are often trade-offs between wildfire risk reduction and other values such as wildlife habitat and recreational use (Copes-Gerbitz et al. 2020; Devisscher et al. 2021), high-intensity thinning will not be appropriate for every stand. However, strategically placed high-intensity thinning treatments, intermixed with low-intensity thinning treatments, have the potential to moderate landscape-scale fire behaviour and effects (Hessburg et al. 2015; Tubbesing et al. 2019).

Pruning after thinning may not be an efficient strategy for reducing potential of passive crown fire

Pruning live and dead branches to reduce ladder fuels is a common wildfire management practice in the western US and Canada (Scott and Reinhardt 2007). Although pruning is generally recommended by government agencies (e.g. California Department of Forestry and Fire Protection 2019; British Columbia Wildfire Service 2023b), there is very little research demonstrating the efficacy, or lack thereof, of pruning (York et al. 2022). For our study area, modelling results indicated that pruning after low- or high-intensity thinning treatments did not substantially increase average live or dead crown base height or reduce potential for passive crown fire. Logistically, it is difficult to prune higher than 3 m above ground using hand tools, which limits the potential additive impacts of pruning. Once smaller trees had been removed through thinning, we found little benefit to pruning the larger residual trees, which generally have higher starting crown base heights. Funding for fuel treatments is limited (Copes-Gerbitz et al. 2020) and pruning is often a substantial proportion of total treatment cost (Erik Leslie, Harrop–Procter Community Forest manager, pers. comm.). Further research on pruning as a fuel reduction technique would help forest managers to prioritise finite resources.

Chipping or pile burning of residue fuel mitigates fire behaviour, but the fire effects associated with chipped fuelbeds remain a concern

Thinning and pruning treatments elevate surface fuel loads when small trees, treetops and branches are scattered over the treatment unit (Schwilk et al. 2009). In the community forests, these residue fuels are commonly ameliorated through pile burning or chipping. Pile burning aims to remove residue fuel from the site, whereas chipping rearranges the residue fuel into a more compact structure (Kreye et al. 2014). Community forest managers are increasingly interested in chipping as a residue fuel management approach, because it is cost effective and does not require follow-up burning (Kreye et al. 2014).

Using models, we found no difference in the change of potential for passive crown fire between pile burning and chipping of residue fuel. As expected, chipping led to a substantial increase in fine fuel load; however, abundant fine fuels were countered by the shallow and compact structure of the chipped fuelbeds. When compact fuelbeds burn, they tend to exhibit low surface fire intensity and flame length, decreasing the potential of crown fire initiation (Glitzenstein et al. 2006; Kreye and Kobziar 2015; Moore et al. 2020).

Community forest managers are additionally concerned with the potential fire effects of alternative residue fuel management approaches. Our modelling results indicated that there was no difference in the change in tree mortality between fuelbeds generated by pile burning or chipping. However, FVS-FFE only models the fire behaviour and effects of the initial flaming front and does not capture potential impacts of long-duration heating (Rebain et al. 2010). In compact fuelbeds, the flaming front is expected to have low intensity and flame length, which has been shown to reduce crown scorch and resultant tree morality (Jerman et al. 2004). However, chipped fuelbeds often experience long-term residual flaming and smouldering, which can cause delayed tree mortality from cambial or root injury (Busse et al. 2005; Knapp et al. 2011; Kreye et al. 2014). Given that FVS-FFE does not capture potential delay, the modelled tree mortality may have been underpredicted for chipped fuelbeds. Further research on the fire effects of chipping would help forest managers better weigh trade-offs between wildfire resistance and treatment cost.

Post-treatment surface fuel load targets were difficult to achieve

Reducing surface fuel load is a common core principle of fuel treatments (Agee and Skinner 2005). A reduction in surface fuel load can lower surface fire intensity and flame length, reducing the chance of crown fire initiation (Van Wagner 1977; Prichard et al. 2010; Stephens et al. 2012). Furthermore, lower fire intensity can prevent trees from being killed by crown scorch or cambium injury (Van Wagner 1973; Ryan and Reinhardt 1988). Following this principle, many of the prescriptions prepared by the community forest managers aimed to reduce fine surface fuel load to below pre-treatment levels. However, our post-treatment field measurements show that fine surface fuel load increased substantially following higher intensity thinning combined with pile burning, demonstrating the difficulty of reducing surface fuel loads due to the influx of fine residue fuel generated by thinning and pruning. In contrast, for the lower-intensity thinning treatments combined with pile burning, fine surface fuel load was similar pre- and post-treatment. Combined, these findings provide important feedback for managers, as they are not always achieving their post-treatment surface fuel load targets.