Dead and down woody debris fuel loads in Canadian forests

Keywords: dead and down woody debris, fire behaviour, fuel load, forest fire, Canada, boreal, surface fuel.

Dead and down woody debris (DWD) is an important component of the forest ecosystem. It forms major structural features important for wildlife habitat (Harmon et al. 1986) and plays a fundamental role in carbon (Woodall et al. 2013) and other key nutrient cycles (Lambert et al. 1980). DWD is a major component of the surface fuel load that contributes to the spread of an active fire (Van Wagner 1977). It is an important contributing factor to forest fire behaviour, particularly as it affects the critical transition phase from surface fire to crown fire. Crown fire threshold is dependent upon the amount of surface fuel consumed within the flame front (flaming combustion), surface fire rate of spread, foliar moisture content and the live crown base height (Van Wagner 1977; Cruz et al. 2004). DWD fuel size and load can directly affect the amount of fuel consumed in a moving surface fire (de Groot et al. 2007; Ottmar 2014; Prichard et al. 2014) and the resulting surface fire intensity (Byram 1959; Alexander 1982). If surface fire intensity reaches the level at which live lower branches of conifer trees are ignited, a point known as the critical surface fire intensity (Van Wagner 1977), a crown fire is initiated. Fire behaviour becomes extreme with the onset of crowning because the fire is exposed to higher wind speeds above the tree canopy (Schroeder and Buck 1970), which increases fire rate of spread, and because total fuel consumption increases due to engagement of crown fuels in the flame front. Both these factors combine to greatly increase overall head fire intensity and fire suppression difficulty (Hirsch and Martell 1996; Alexander and Cruz 2019). When a surface fire becomes a crown fire, head fire intensity can increase by an order of magnitude. Spreading surface fires typically have a head fire intensity of 500–4000 kW/m, whereas crown fires can have 4000–90 000+ kW/m (Alexander and de Groot 1988; Alexander and Lanoville 1989; Stocks et al. 2004).

DWD fuel size and load varies greatly both spatially (Keane et al. 2012) and temporally (Hély et al. 2000) and has many contributing factors. The accumulation of DWD on the forest floor is a balance between tree mortality and breakage versus decomposition (Harmon et al. 1986) that changes over the life history of the stand (Woodall et al. 2013). It varies depending on the species composition (Bernier et al. 2007), disturbance history (Stocks 1987; Keane et al. 2012) and site productivity (Woodall et al. 2013). DWD increases due to mortality during life cycle processes such as natural thinning as immature stands grow, and stand break-up of over-mature stands. It also increases due to periodic natural disturbances including wind throw, insect outbreaks and snow damage (Brown and See 1981). DWD decomposition, on the other hand, is influenced by patterns of temperature, moisture and substrate quality (Harmon et al. 1986). The spatial distribution of DWD varies by size classes where coarse woody debris tends to be close to tree boles and decomposes slowly compared with finer woody debris that is much lighter and more easily dispersed creating a more uniform pattern (Keane et al. 2012). Another important agent of DWD reduction is fire, which can consume a great proportion of DWD depending on fire intensity and fire weather conditions (Forestry Canada Fire Danger Group 1992).

As mentioned earlier, DWD fuel load and the amount consumed by fire directly affect fire intensity (Byram 1959), and the rate of combustion influences emissions (Cofer et al. 1990). These variables are influenced by DWD size class, moisture content and combustion phase (i.e. flaming, smouldering and residual), although most models do not currently differentiate between the latter two phases (Ottmar 2014). Quantification of DWD by size class is important as the surface to volume ratio affects drying rates and therefore ignition and consumption. Smaller diameter fuels require less heat to ignite and are generally consumed faster. DWD fuel load is defined as the total dry weight biomass per unit area (Keane 2013) and is an input in fire behaviour prediction (e.g. Albini 1976), effects (Reinhardt et al. 1997; de Groot 2010) and carbon emission models (Kurz et al. 2009). To date, DWD fuels have not been modelled explicitly in the Canadian Forest Fire Behaviour Prediction (FBP) System (Forestry Canada Fire Danger Group 1992), one of the major subsystems in the Canadian Forest Fire Danger Rating System (CFFDRS) (Stocks et al. 1989), which is the primary source of daily fire intelligence in Canada. Instead, DWD is combined with forest floor and understorey vegetation fuels to represent the entire surface layer as a single fuel component, due to the capabilities of model development at the time. However, advances have been made since then to separate surface fuels into forest floor and DWD fuel layers, each with multiple components (Amiro et al. 2009; de Groot et al. 2009, 2013). Specifically, DWD is modelled explicitly as a distinct fuel layer comprised of individual diameter size-classes in the Canadian Fire Effects Model (CanFIRE; de Groot 2010), which has shown significant improvement in fire behaviour predictions (de Groot et al. 2013) and carbon emission assessment (de Groot et al. 2007) owing to these changes.

Standard fuel models have been developed for fire behaviour prediction in Canada using a classification system of 16 fuel types in the FBP System (Forestry Canada Fire Danger Group 1992), which are based on dominant forest vegetation types found across Canada. The FBP System calculates fire behaviour attributes including fuel consumption, rate of spread and head fire intensity for each FBP System Fuel Type. These models are driven primarily by components of the Canadian Forest Fire Weather Index (FWI) System (Van Wagner 1987), representing the effect of fire weather on dead fuel moisture content and fire behaviour.

FBP System standard fuel models have components of surface and canopy fuels, which were assigned with an average fuel load to each component for every fuel type. As part of a project to build the next generation CFFDRS (CFFDRS-2025), research efforts to update the FBP System include adding the ability to quantitatively represent fuel structural components, which allows surface and ground fuels to be separated into individual fuel layers (i.e. litter, duff, DWD) (Canadian Forest Service Fire Danger Group 2021). This greatly increases ground and surface fuel model flexibility to more accurately reflect the diverse range of surface fuel conditions in Canada. Similar changes are planned for the aboveground tree fuel component, which will greatly expand the range of standing timber fuel types that are represented in the FBP System, including special cases such as the damaged or disturbed forests with increased DWD fuel loads. By incorporating an adjustable fuel load dimension for individual layers, like DWD, fire behaviour prediction becomes more robust and dynamic. In order to implement these changes in the next generation FBP System and extension models like CanFIRE, reliable DWD fuel load information representing forest stands across the country is required. Currently, neither a national DWD fuels database nor DWD fuel load models are available in Canada.

Assessing wildland surface fuels in general is difficult (Keane 2013), especially when DWD fuels are considered due to their high degree of spatial variability over relatively small areas (Gould et al. 2008; Woodall and Liknes 2008). Regression models predicting DWD fuel loads have been established based on species, age, temperature and moisture regimes, and disturbance, but they are limited to selected forest types or regions in Canada (e.g. Lee et al. 1997; Hély et al. 2000; Pedlar et al. 2002; Ter-Mikaelian et al. 2008; Allard and Park 2013). The objectives of this study were to (1) assemble existing Canadian datasets to characterise DWD fuel load, (2) examine the geographical variation of DWD, and (3) analyse potential factors to predict DWD fuel loads across the country.

Numerous Canadian datasets were examined to identify those containing DWD information as well as information on dominant tree species, ecozone and/or spatial coordinates of the sampling plots. There were five datasets that met these criteria for inclusion in this study (Table 1): the National Forest Inventory (NFI), the Forest Ecosystem Carbon Database (FECD), the FBP System experimental fire database (FIRE dataset hereafter), the Energy from the Forest Program of Canadian Forest Service (ENFOR) dataset, and the Canada-Ontario Forest Resource Development Agreement (COFRDA) forest fuels database. All datasets fall within the predominately forested landmass of Canada as defined by the Ecological Stratification Working Group (ESWG 1995) (Fig. 1).

Table 1.  Number of plots (n) by dataset
DWD size class and explanatory variables for Database I are unshaded. Those explanatory variables shaded in grey are included in Database II, in addition to all those from Database I. Diameter size classes are as follows: fine woody debris (FWD) <1 cm, medium woody debris (MWD) 1–7 cm, coarse woody debris (CWD) >7 cm, and total DWD includes all size classes. NFI, National Forest Inventory; FECS, Forest Ecosystem Carbon Database; FIRE, Fire Behaviour Prediction System experimental database; ENFOR, Energy from the Forest Program; COFRDA, Canada-Ontario Forest Resource Development Agreement

Fig. 1.  Map of plot locations in the DWD fuel load dataset across ecozones. The datasets include the NFI, FECD, FIRE, ENFOR and COFRDA.

The NFI is a collection of national permanent sample plots across Canadian forests (Gillis et al. 2005) maintained by the Canadian Forest Service. Attributes are obtained from ground plots, photo plots and remote sensing data. DWD of various size classes were sampled from the ground plots using both transects and microplots (Table 2). The NFI is the most complete dataset within the study area containing 766 plots across the country that met the study requirements (Fig. 1); latitude, longitude, stand age, drainage class, tree species, diameter at breast height and stand density (both live and dead standing) were also measured (Table 1). In this dataset, drainage class is a measure of how quickly water is drained from the soil ranging from very rapid, rapid, well, moderately well, imperfect, poor, to very poor (Ontario Institute of Pedology 1985). DWD fuel loads were measured for three size classes as fine (<1.0 cm), medium (1.0–7.5 cm), coarse (>7.5 cm) and total woody debris (all sizes) (Table 2).

Table 2.  Datasets complied for the study, including their source, field method for collection and DWD size class distributions
NFI, National Forest Inventory; FECS, Forest Ecosystem Carbon Database; FIRE, Fire Behaviour Prediction System experimental database; ENFOR, Energy from the Forest Program; COFRDA, Canada-Ontario Forest Resource Development Agreement

The FECD was part of a series of projects designed to address the data needs of modellers studying the impacts of changing climate and site characteristics on forest carbon stocks and fluxes (Shaw et al. 2005). The FECD focussed on compiling existing plot-level data to be used in the estimate of aboveground and belowground carbon pools. A total of 706 plots from eight sources were included in the dataset collected from 1986 to 2002 (Fig. 1). Only 94 plots met the minimum required attributes of this study. The dataset includes latitude, longitude, dominant tree species, stand age and drainage class (Table 1). DWD fuel loads were collected for two size classes in this database, less than all other datasets (Table 2).

Fuel load data in the FIRE dataset were collected on six experimental burning projects conducted by the Canadian Forest Service over 40 years to build the FBP System (Forestry Canada Fire Danger Group 1992) (Fig. 1). These experimental burning projects were conducted at Hondo (Quintilio et al. 1991) and Darwin Lake (Quintilio et al. 1977) in Alberta; Sharpsand Creek (Stocks 1987) and Kenshoe Lake (Stocks 1989) in Ontario; and Fort Providence (Stocks et al. 2004) and Porter Lake (Alexander et al. 1991) in the North-west Territories (Fig. 1). DWD fuel loads in this dataset were measured by multiple size classes (Table 2) but included a limited number of attribute variables (Table 1).

Nalder et al. (1997, 1999) and others¹ created the ENFOR dataset for DWD fuels in the boreal forests of western and northern Canada (Alberta and North-west Territories; Fig. 1). Fuel loads were measured for multiple size classes (Table 2) and included information by dominant tree species and stand age (Table 1).

Lastly, the COFRDA forest fuels dataset is a detailed inventory of close to 300 stands across the Boreal and Great Lakes–St. Lawrence forest regions of northern Ontario (Stocks et al. 1990) (Fig. 1). DWD fuel loads were measured in three size classes at each plot with dominant tree species information of the plot (Table 2).

Two fuels databases were compiled to characterise Canadian DWD fuel loads, and to analyse potential factors predicting DWD fuel loads across the country (objectives 1 and 2). One DWD fuels database contained the plot-level data from all five different datasets (Table 1) (Database I), and the second was a subset containing only the NFI dataset (Database II). Although Database I was larger, because it included all five DWD datasets, the number of variables was limited by data not available for drainage and stand structure characteristics in some DWD datasets. In contrast, Database II contained only the NFI dataset, which was smaller than Database I, but it included a large amount of data for all variables of interest (i.e. structural attributes). Database I and Database II were both national in scope. Although Database II included a more extensive list of variables than Database I and it contained more than half (59.6%) of all available DWD plot data (Table 1), Database II was still lacking in data for certain species or species groups. Therefore, Database I was included in this study to provide analysis over a broader range of species with a larger dataset.

All DWD data in this study was collected using the line intersect method (Van Wagner 1968). When compiling the databases, DWD fuel load and dominant tree species were the primary attributes of interest. In order to be consistent, the DWD data were grouped into three diameter size classes: fine (0–0.99 cm) (FWD), medium (1–7 cm) (MWD) and coarse (7+ cm) (CWD). The diameter distribution of the NFI plots was slightly different for the top end of MWD (i.e. 7.5 cm) and lower end of CWD (Table 2). The FECD dataset did not have a separate FWD class and COFRDA sampled FWD up to 2.99 cm (Table 2); these were aggregated into the MWD class. The assumption was that these small discrepancies would have minimal influence on the final fuel loading outcomes (Hollis et al. 2010). Total DWD fuel load was calculated as the total fuel load of all size classes. There were some instances where zero fuel load was measured for a size class, which usually occurred in the FWD size class, although zero fuel load was not a common occurrence in the dataset. In the case of no entry for a given fuel size class, the record was removed from the analysis.

Plots from all sources were assigned to the ecozone level of the national ecological framework (ESWG 1995) (Fig. 1). For this study, only predominately forested ecozones were included and the Boreal Shield was split into east and west subzones (Stocks et al. 2003). Hudson Plains and Taiga Shield were excluded due to a lack of plots, resulting in a total of nine ecozones represented. In addition to the sampled plot-level covariates of tree species, soil drainage and stand age, 36 bioclimatic variables (McKenney et al. 2013), longitude and elevation were extracted for each plot based on spatial coordinates, to provide additional explanatory power. An initial exploratory analysis using only the sampled plot-level covariates were all found to have a significant effect on DWD fuel loads based on ANOVA and regression analysis. However, the best models using only plot-level covariates could only explain 30–40% of the variance for all DWD size classes. In order to improve the prediction power of these models, additional bioclimate variables were added to the analysis to expand on the climatic differences within the ecozone delineations. The majority of the initial 36 bioclimatic variables were highly correlated with each other based on Pearson’s correlation coefficient (r ≥ 0.7; Dormann et al. 2013; Blouin et al. 2016). As a result, only nine bioclimatic variables were selected for the final analysis where r < 0.7 (Table 3). In addition, species were grouped into genera to increase sample sizes for less common species (Tables S1–S4, Supplementary Material) with limited spatial extent across the country, that is, maple (Acer), birch (Betula), Douglas-fir (Pseudotsuga) and hemlock (Tsuga).

Table 3.  Uncorrelated standard bioclimatic variables produced with Canadian and North American surfaces by McKenney et al. (2013) (http://cfs.nrcan.gc.ca/projects/3. See https://fennerschool.anu.edu.au/research/products/anuclim for information about ANUCLIM.)

Based on Database II, one-way ANOVA was used to compare the DWD fuel loads by ecozone and DWD size classes across the study area. We used Database II for the comparison because the sampling plots in the NFI database were designed to represent various forest ecosystems, which may increase the reliability of the comparisons. To compensate for the skewed nature of the data to low fuel loads, the data were square root transformed. We also compared DWD fuel loads by DWD size classes for forest stands dominated by various species and species groups across the country.

We developed the predictive models of DWD at various sizes with all sampling plots from Database I. Explanatory variables including age and drainage class in addition to elevation, longitude and bioclimatic variables (Table 3) were considered in these models when available (Table 1). For its simplicity, a multiple linear regression model was carried out on all uncorrelated variables by size class for the entire database and for each dominant species or genus group when sample size was limited, respectively. An R package leaps was used for model selection by conducting an exhaustive search of the minimum number of variables that resulted in the simplest model using main factors with the optimal adjusted R² (P < 0.05). By checking the residual plot of each model, a quadratic term of the predictors was tested in the modelling procedure. Proportional contribution of each variable was calculated based on the proportional variation explained by the variable to the total.

For species or species groups where no statistically significant models could be built based on Database I, the same multiple linear regression analysis was performed using Database II in order to use the additional plot-level structural variables to improve the goodness of fit. We built the regression models by species level for the whole study area and for each ecozone (which was shown to be significant based on ANOVA results) where plot numbers were sufficient. To compensate for the reduction in the number of plots, a basic rule of thumb was followed where analysis was not performed with fewer than seven plots and a minimum of one explanatory variable could be included for every five plots, which is suggested as the minimum number of records necessary to build a reliable model (Tabachnick and Fidell 2019).

The largest average TWD (total woody debris) fuel loads were found in western Canada in the Pacific Maritime and Montane Cordillera ecozones (Fig. 2). Based on our exploratory analysis, ecozone was found to be a significant variable explaining up to 85% of the variability in fuel load for some dominant tree species and all size classes. The lowest TWD fuel loads were found in the far northern ecozone of the Taiga Plains and more southern Mixedwood Plains, which is dominated by more temperate tree species. Mean TWD is relatively consistent across the boreal ecozones into the Atlantic Maritime (Fig. 2), although it is important to note that there is a lot of variability in fuel loads for all size classes, where standard deviations are often equal to or greater than the mean fuel load (Tables S1–S4, Supplementary Material). CWD makes up the greatest proportion of TWD fuel load in all ecozones (Fig. 2). The distribution of other DWD size classes varied among ecozones where the Atlantic Maritime had the lowest values for FWD and the Pacific Maritime, Montane Cordillera and Boreal Cordillera had the lowest proportional MWD fuel loads (Fig. 2).

Fig. 2.  Average TWD (including all diameter size classes) by ecozone with the pie charts of proportional woody debris by size class; FWD (<1 cm), MWD (1–7 cm) and CWD (>7 cm). Ecozones are as follows: BC (Boreal Cordillera), PM (Pacific Maritime), MC (Montane Cordillera), TP (Taiga Plains), BP (Boreal Plains), BSW (Boreal Shield West), BSE (Boreal Shield East), MP (Mixedwood Plains) and AM (Atlantic Maritime).

When grouped by genera, median FWD fuel loads were all under 0.5 kg/m² (Fig. 3), with the exception of a few outliers. FWD fuel loads were greatest for coastal Douglas-fir (Pseudotsuga menziesii var. menziesii) and hemlock (Tsuga), due to the high FWD loads of Douglas-fir (mean 0.68 kg/m²) and western hemlock (Tsuga heterophylla) (mean 0.44 kg/m²) in the Pacific Maritime ecozone (Table S1). MWD loads were higher than FWD loads, upwards of 1 kg/m² (Fig. 3), again dominated by large western coniferous tree species including both Douglas-fir and western hemlock (Table S2). CWD fuel loads were very low for larch (Larix), mean 0.26 kg/m² (Table S3, Fig. 3), and under 1 kg/m² on average for sugar maple (Acer saccharum), eastern white pine (Pinus strobus) and balsam poplar (Populus balsamifera). TWD fuel loads were greatest for hemlock (Tsuga) (Fig. 3), dominated by coastal western hemlock (Table S4).

Fig. 3.  Box plot comparisons of fuel loading by size class (FWD, MWD, CWD and TWD) by genus.

Multiple linear regressions fitted by genus, using Database I, resulted in significant models for six of the genera tested (Table S5, Supplementary Material) explaining 35–75% of the fuel load variability (Table 4). Of the initial 14 explanatory variables considered, the most commonly significant were temperature-based variables (i.e. annual mean temperature, isothermality, mean diurnal temperature or mean temperature of the wettest quarter), but they rarely explained more than 30% of the variability in DWD fuel load. Age and soil drainage were not as often found to be significant but when they were significant could explain up to 50% of the variability (i.e. MWD for Douglas-fir and hemlock, which are high drought and water tolerant species, respectively). The same explanatory variables were often significant across different size classes within the same genera (Table 4).

Table 4.  Summary of multiple linear regression models by genus using Database I, including significant explanatory variables and their estimated coefficients, proportional variability explained (Exp Var) and adjusted (Adj) R²
T, temperature; P, precipitation; Q, quarter; SD, soil drainage (categorical); µ, mean; DBH, diameter at breast height. Only significant (P < 0.05) models are included

No significant models were found for larch and for the three most widespread genera found in Canada, spruce (Picea), pine (Pinus) and poplar/aspen (Populus) based on Database I (Table 4). Models fitted with additional structural explanatory variables, DBH and stand density (both live and dead) using Database II showed significant improvement (Table 5 and Table S6, Supplementary Material) for pine only. Models for pine CWD and TWD were significant (P < 0.05) with adjusted-R² of 0.37 and 0.35, respectively (Table S6). In both cases elevation and DBH were the only significant explanatory variables included in the final model, where most of the variability was explained by elevation differences (Table 5).

Table 5.  Summary of multiple linear regression models by genus using Database II, including significant explanatory variables and their estimated coefficients, explained variability (Exp Var) and adjusted (Adj) R²
DBH, diameter at breast height. Only significant (P < 0.05) models are included

Additional variability in common boreal genera could be explained by separate dominant tree species (Table 6 and Table S7, Supplementary Material), except for those most commonly found across the country (which is addressed in the following paragraph). Although the number of plots were reduced, models for many DWD size classes were improved. For example, the explained variability for the spruce genus was 30.3% of TWD fuel load (Table 6), but separating by species the explained variability in TWD fuel load for Engelmann spruce (Picea engelmannii) increased to 55.1% (Table 6). Similar examples can be seen for spruce and poplar/aspen (Table S7 and Table 6). Significant models were found for all size classes for eastern white pine (Pinus strobus), where the density of standing live trees was the only covariate that remained for MWD, CWD and TWD and explained greater than 50% of the variability in fuel load for each size class (Table 6). Explanatory variables that related to temperature were again the most commonly selected in the regression models, although often did not contribute much in explaining the variability. When included, age and structural variables (in Database II), DBH and density of standing live trees were more important. Significant models that explained greater than 30% of the variability were found for seven of the nine dominant tree species in just under half of the size class/species combinations.

Table 6.  Summary of multiple linear regression models by dominant species using Database II, including significant explanatory variables and their estimated coefficients, explained variability (Exp Var) and adjusted (Adj) R²
T, temperature; P, precipitation; Q, quarter; SD, soil drainage (categorical); µ, mean; DBH, diameter at breast height; DSD, density standing dead; DSL, density standing live. Only significant (P < 0.05) models are included

For the remaining two common boreal tree species for which no significant models could be found for any size class (Jack pine (Pinus banksiana) and black spruce (Picea mariana)), as well as the two species for which only one size class was available to develop prediction models (white spruce (Picea glauca) and trembling aspen (Populus tremuloides)), higher levels of explained variability in DWD fuel loads were achieved when the prediction models were developed by ecozone, that is, where the number of plots permitted this (Table 7 and Table S8, Supplementary Material). This was possible for black spruce, white spruce, Jack pine and trembling aspen. Significant models that explained a good proportion of fuel load variability were found within each species for each ecozone, with the exception of black spruce in the Boreal Shield East (Table 7). The key explanatory variables in the models were again more similar between DWD size classes than between ecozones within a species or between species. These included age, temperature, longitude and structural variables (DBH, DSD) as well as soil drainage, which were most important based on high proportions of variability explained in the resultant models.

Table 7.  Summary of multiple linear regression models by ecozone (Figs 1 and 2) and species using Database II, including significant explanatory variables and their estimated coefficients, explained variability (Exp Var) and adjusted (Adj) R²
T, temperature; P, precipitation; Q, quarter; SD, soil drainage (categorical); µ, mean; DBH, diameter at breast height; DSD, density standing dead; DSL, density standing live. Only significant (P < 0.05) models are included

This study represents the first research in Canada to develop DWD fuel load models for most dominant forest types across the country. Because it is difficult and expensive to monitor and sample, DWD fuel load data is typically sparse. The national datasets presented herein are themselves of great value, particularly to fire managers across the country (see Table S9 in the Supplementary Material to crosswalk results to the current benchmark FBP System Fuel Types). Recent studies on sensitivity analysis have shown that variations in fuel load can have significant impact on fuel consumption and therefore fire behaviour and emission predictions (Kennedy et al. 2020). Any information that allows modellers to tailor fuel load estimates is helpful. Using regression modelling techniques, we developed models with high predictive power for 80% of the dominant Canadian forest types/DWD size classes, which improves our ability to characterise DWD fuels for improved fire behaviour, fire effects and fire emission models used within research and operational fire management (see Table S9 in the Supplementary Material for general guidelines on how results can be applied). Where explanatory variables are not available, mean and standard deviation values of DWD by species and ecozone provide a much needed starting point to calibrate models.

The databases analysed in this study revealed a wide range in DWD fuel loads, as also shown in other studies (Keane 2016). As previously mentioned, species and climatic gradients strongly influence DWD production (Bernier et al. 2007), accumulation (Allard and Park 2013) and decomposition (Harmon et al. 2000). DWD variability across climatic gradients in this study are similar to findings by Woodall et al. (2013), which showed greatest loadings in the Pacific North West of the USA, an area of cool wet climate and infrequent fire very similar to Canada’s Pacific Maritime ecozone. Bioclimatic variables, especially those derived from temperature, were significant in almost all predictive models for all DWD size classes. Although, these were often not as important as variables such as age, drainage, elevation or structural attributes when included. Overall, there was more consistency in terms of explanatory variable selections between size classes than between genera or species. For some genera, further division into species groups was necessary to achieve a better fitted model, which may relate to greater differences in tree morphology within some genera, that is, white spruce and black spruce have very different structural growth patterns and occupy different niches. Species has been found to be a significant variable also in other models of DWD (Hély et al. 2000; Allard and Park 2013).

Drainage class and age have also been shown to influence DWD load, particularly for CWD (Gould et al. 2008; Allard and Park 2013). Although these were significant variables in less than half of the regression models fitted in this study, when age was a significant variable, it could in some cases explain over half of the variability (i.e. FWD Engelmann spruce and MWD hemlock). We assume that drainage class may show significant influence on DWD when the species are more tolerant to soil moisture changes, because drainage class may not vary much for species with less tolerance to soil moisture regime changes. The lack of consistency of age influencing DWD has also been found elsewhere (Ter-Mikaelian et al. 2008), whereas others have found disturbance, which can be synonymous with age, to be important (Hély et al. 2000; Moroni 2006). It is not clear why age has not been consistent in predicting DWD; however, we found that age appeared to be significant only to species that are adapted to mesic or drier habitats (e.g. Jack pine), not to species that are adapted to wetter habitats (e.g. black spruce). Furthermore, a lack of age data for all plots may be the other reason that age did not show significant influence in some models.

Summarising DWD fuel loading is complicated, and additional factors including small scale disturbance events and silvicultural practices further confound national scale summaries (Woodall and Liknes 2008). Disturbance patterns that occur at the stand level can strongly influence DWD accumulation rates (Moroni 2006), which may result in weak DWD relationships to landscape scale ecosystem characteristics like species type and climate (Brown and See 1981). This may be why some forest types required classification to the ecozone level to capture added variability due to either the widespread distribution of some species or greater influence of stand scale processes across Canadian forested lands.

With the exception of the Pacific Maritime ecozone (which has some DWD fuel loads >11 kg/m²), average Canadian TWD fuel loads are low (<6 kg/m², with most ecozones having <3 kg/m²) in comparison to typical fuel loads in the forest floor (5–20 kg/m², Letang and de Groot 2012) and standing timber (5–16 kg/m², Penner et al. 1997). However, DWD fuels are important because they are an influential factor on fuel consumption, surface fire intensity and potential fire behaviour overall. A high proportion of FWD and MWD (and a small proportion of CWD) are consumed by surface fire during the flaming combustion phase (Hollis et al. 2010; Prichard et al. 2014), directly contributing to surface fire intensity. In contrast, the organic forest floor fuel layer, which is the other primary fuel stratum contributing to surface fire intensity, has a very different pattern. Forest floor fuel loads are much higher than DWD and consumption values can be high, but the amount consumed is wide-ranging and dependent on the level of long-term drying (de Groot et al. 2009). However, even when forest floor fuel consumption is very high, only the surface layer of litter, lichen and mosses is consumed at a high rate in the flaming combustion phase (90–95%), and very little of the subsurface duff fuels (F and H layers) are consumed during flaming combustion (10% for upper duff, 0% for lower duff) (Prichard et al. 2007). As a result, only a small amount of the total organic soil layer burned contributes to surface fire intensity. By separating these strata in the next generation CFFDRS fuels (Canadian Forest Service Fire Danger Group 2021), the contribution of DWD to overall surface fuel loads becomes more important. These data are a starting point to help initialise fuel loads in resultant DWD consumption models currently in development.

As with most scientific analyses this study was not without limitations. Spatially, many of the sample plots, particularly in more southern parts of the country were clustered, with fewer plots in more northern areas; there were no plots in the Taiga Shield and Hudson Plains ecozones. Some genera had fewer than 40 plots across the country including maple, birch, larch, Douglas-fir, hemlock and cedar, and low plot numbers restricted the analysis for these forest types. In addition, structural variables of DBH and stem density (live or dead) were only available in Database II and may have explained further variability in the models created using Database I. Consistency in data collection standards may also influence results, that is, using the same size classes may have contributed to higher variability within some of the regression models. Future field studies should use consistent size classes and collect more in-situ data like age, disturbance regimes (i.e. time since fire) and structural attributes. Future research could also use other forms of modelling techniques such as non-linear models or machine learning, although multiple linear regression did have good results in most cases as shown in our study and allowed exploration of explanatory variables. The compiled databases provide a much needed starting point for eventual mapping of DWD that could be done with more sophisticated machine learning techniques and combined with remote sensing information like Lidar (light detection and ranging) across the landscape to obtain more plot level data (Keane et al. 2012; Queiroz et al. 2019).

DWD fuel loads in natural forest stands across Canada are highly variable. The relationship between bioclimatic variables, elevation, drainage, age and tree structure with woody debris fuel load is complicated, and DWD fuel loads therefore cannot be described/quantified nationally based on a simple set of variables. The in-situ measured variables, dominant tree species and ecozone, were significant predictive factors but only explain 30% of variation on their own. Age and drainage class were also significant factors but only increase explained variance by an additional 1–9%, depending on DWD size class. Addition of bioclimatic variables and structural attributes to the models improved models in most cases. Overall, the models and datasets developed in this study capture a good portion of the inherent variability of DWD fuel loads, which provides a first approximation of DWD fuel loads in standing timber fuel types across Canada. Based on this study, the fuel loads are consistent with other regional scale analyses across Canada and broad scale patterns are similar to other national analyses. These results together with forest floor fuel loads from Letang and de Groot (2012) give a thorough picture of surface and ground dead fuel loading across Canada, with only the live understorey vegetation component missing to complete a national surface fuel loading database required for wildland fire modelling and implementation of planned changes to the next generation CFFDRS (Canadian Forest Service Fire Danger Group 2021). These data and models provide a foundation on which to build quantitative fuel load and consumption models for improved prediction of fire behaviour, fire effects and forest carbon dynamics.

The authors declare no conflicts of interest.

This research did not receive any specific funding.