1. Introduction
Wildfires are the most extensive stand-initiating disturbance in the northwestern Canadian boreal forest, typically recurring every 50–100 years [
1,
2]. When wildfires occur, they burn with varying intensities (energy release) in response to fire weather, topography, and fuel type, producing a range of burn severities. Burn severity is defined as the ecological impacts of fire on vegetation and soils [
3,
4]. Many boreal forest plants have adapted to repeated wildfires though traits such as resprouting or suckering, seed banking, or, in the case of some tree species, serotiny. Serotinous and semi-serotinous conifer tree species have cones that may open in response to and survive some heating, and retain some viable seeds in the canopy following wildfires. Through this mechanism, serotinous species can produce extensive seed rains from aerial seedbanks immediately following fire [
1,
5]. Wildfire burn severity has important implications for post-fire understory vegetation communities and recruitment of seedlings. Heating and combustion from wildfires kill some trees and may reduce the viability of seeds in aerial seedbanks (including those of serotinous species) beyond a threshold of fire intensity or if the duration of heating is extensive [
6,
7]. Variable combustion of organic soils provides diverse seedbeds for plants and trees, ranging from thick remnant organic layers to exposed mineral soils, and alters the composition and exposure of post-fire soil seed banks [
8,
9]. Some burning of organic soils promotes vegetative regeneration, but deep burning may damage roots and rhizomes, negatively affecting the capacity of resprouting species to regenerate following fires [
5].
In many ecosystems, burn severity is a dominant and enduring control on post-fire understory vegetation assemblies [
10,
11,
12] and seedling recruitment [
11,
13,
14], influencing the resulting structure and composition of forests. Although burned sites in the boreal forest generally return to a mature forested stand structure within 100 years [
15], researchers using remote sensing to examine the post-fire recovery of vegetation following wildfires have found different rates of revegetation amongst burn severity classes. Severely burned sites demonstrated the highest decline in vegetation immediately post-fire [
16,
17]. In the years following a wildfire, severely burned sites subsequently experienced the largest increases in vegetation, indicating either forest recovery or colonization of these sites by disturbance-favouring plants and trees [
16,
17]. In North American boreal forests, post-fire understory vegetation communities in black spruce (
Picea mariana (Mill.) BSP.) [
18], jack pine (
Pinus banksiana Lamb.) [
19], and mixed broadleaf and coniferous stands [
20] are influenced by surface burn severity and depth of burn, in conjunction with the availability of seed sources and vegetative propagules. In these studies, colonizing species such as graminoids and annual forbs established themselves broadly in severely burned areas, whereas slow-growing lichens, evergreen shrubs, and higher species richness were more prominent in low severity and scorched areas [
18,
19,
20]. Lower densities of recruitment of coniferous trees have been observed when sites burned severely and at short intervals [
21,
22], and increased proportions of early-successional tree species, such as jack pine and trembling aspen (
Populus tremuloides Michx.) are associated with high severity burning [
23,
24]. The relative dominance of different species of trees and the density of post-fire forests are lasting legacies of boreal wildfire severity [
21,
23,
25].
When burn severity is studied at a broader landscape scale, that is, across multiple forest types and wildfires, the effects of burn severity on post-fire vegetation communities and recruitment may be challenging to detect. Burn severity is correlated to pre-fire forest type and stand structure [
8,
26,
27,
28], potentially obscuring or explaining observed effects of burn severity on post-fire plants and trees. Studies of burn severity that encompass multiple forest types have identified topoedaphic and pre-fire forest conditions as the primary post-fire drivers of understory plant communities and site suitability for tree species [
18,
29,
30,
31,
32], leading some researchers to characterize burn severity as a secondary “filtering” effect beneath the dominant landscape and climatological controls.
Ranges of burn severity and the relatively infrequent occurrence of large wildfires (≥200 ha) produce a mosaic of stand ages and patterns on the landscape, in regions with mixed- and high-severity fire regimes [
33,
34,
35]. Wildfires interact with past burns, as previous fires and burn severity determine current fuels. Abnormally short fire frequencies are implicated in the dominance shifts of tree species [
36], low stocking in post-fire forests [
22], and even near-deforestation [
37], with implications for forest resilience [
38]. Furthermore, burn severity interacts with fire frequency, potentially reinforcing vegetation type conversions [
39]. Wildfires are a weather and, therefore, climate-driven disturbance. Fires are expected to increase in size, frequency, and intensity (and therefore in severity) [
40,
41,
42] in North America as the climate warms and severe fire weather increases [
43]. The forests of the Canadian Northwest Territories provide an interesting opportunity to study the effects of extensive free-burning wildfires in an ecosystem with multiple dominant coniferous and broadleaf tree species, across a moisture gradient ranging from hydric to xeric. Given the ecologically important role and actively changing patterns of fire in the boreal forest, studies characterizing the relative importance of fire effects and fire history, and non-fire and climate drivers in determining post-fire vegetation assemblies and species composition shifts will provide insights into the trajectories of future forests.
This study describes post-fire vegetation communities and seedling recruitment across a broad range of topoedaphic vegetation classes and levels of burn severity, to identify direct and indirect drivers of these assemblies in the northwestern Canadian boreal forest. In support of this goal, our objectives were: 1. To characterize post-fire vegetation assemblies and recruitment of seedlings across burn severity and topoedaphic gradients; 2. To assess the relative importance of climate and pre-fire forests, burn severity and fire history, and post-fire soils to understory vegetation communities and shifts in the dominance of tree species; and 3. To identify direct and indirect effects of fires on post-fire vegetation, as well as drivers of shifts in the dominance of tree species in the post-fire cohort.
3. Results
A broad range of burn severity was represented in the field sites (
Appendix A:
Figure A2). The BSI values of field sites ranged from 0.5 to 4, the CFSI values from 0 to 6, and percent overstory mortality ranged from 6.25% to 100% [
28]. The surface (BSI) and overstory (CFSI) burn severity were statistically related to topoedaphic vegetation classes (ANOVA, ***
p < 0.001 and *
p = 0.02, respectively;
Table 4), but overstory mortality was not. Post-hoc comparisons of least-squares means with a Tukey test confirmed some statistical differences in burn severity amongst topoedaphic vegetation classes for BSI (*
p ≤ 0.05;
Appendix A:
Figure A2). Surface burn severity was lowest in open wetlands and highest in jack pine uplands. All other topoedaphic vegetation classes had BSI values that were similar to one of these two groups. The differences in least-squares means of CFSI between topoedaphic vegetation classes were not significant at α = 0.05 (
Appendix A:
Figure A2).
The richness of the understory vascular plant species sampled at the field sites ranged from three to 20. Both understory vegetation community diversity and seedling density were statistically related to topoedaphic vegetation classes (
Table 4;
Appendix A:
Figure A2). Interactions between TSLF and topoedaphic vegetation classes, and BSI and topoedaphic vegetation classes significantly explained the variability in understory plant diversity (Type III ANOVA, *
p ≤ 0.02;
Table 4). The density of seedlings was significantly explained by both topoedaphic vegetation classes and BSI (Type II ANOVA,
p < 0.04), but not by TSLF or CFSI, or by the interactions between topoedaphic vegetation classes and these two metrics (
Table 4;
Appendix A:
Figure A2). All sites with zero seedling establishment were open wetlands (
n = 7). Of those sites that experienced some regeneration, seedlings ha
−1 ranged from 25 to >75,000. The natural logarithm of the density of seedlings was significantly lower in open wetlands, treed wetlands, and upland spruce, and higher in upland jack pine and upland mixedwood topoedaphic vegetation classes (comparison of least-squares means, with a Tukey test, α = 0.05). The post-fire seedling density was statistically greater in sites that experienced >17 years between fires (Wilcoxon signed-rank test, *
p = 0.02).
The two-dimensional NMDS of understory vascular plant communities had a stress of 0.20 (
Figure 2). Similarity of understory species communities was primarily related to the physicochemical properties of the soil; however, pre-fire forest structural characteristics of basal area and absolute stem density of overstory and understory trees were also influential (
Figure 2). BSI was also statistically related to understory species community dissimilarity, and TSLF was nearly significant (
p = 0.052, 999 permutations). Although soil properties were explained by topoedaphic vegetation classes; the organic soil depth, total nitrogen, total carbon, potassium, calcium, and magnesium were also statistically (Type II ANOVA; α = 0.05) related to BSI when controlling for the effect of topoedaphic vegetation class. Therefore, some soil properties were affected by fire (
Appendix A:
Table A1). Topoedaphic vegetation classes tended to occupy characteristic areas of ordination space, but there was some overlap between the normal confidence ellipses of classes. Upland mixedwood and upland jack pine groups were especially intermingled (
Figure 2a), and mixedwood communities occurred in a sub-region of the broader environmental space occupied by jack pine. Similar patterns are identifiable in the post-fire understory indicator species of each topoedaphic vegetation class (
Table 5). All topoedaphic vegetation classes had unique significant indicator species, with the exception of jack pine uplands, which shared all significant indicator species with the upland mixedwood group, and some with the upland spruce group (
Table 5).
Potentilla palustris (L.) Scop.,
Betula glandulosa Michx.,
Epilobium palustre L., and
Myrica gale L. were unique indicator species of open wetlands. Treed wetlands had unique indicator species of
Rubus chamaemorus L.,
Vaccinium caespitosum Michx., and
Vaccinium oxycoccos L.
Viburnum edule (Michx.) Raf. was a unique indicator species of upland mixedwood sites. Furthermore, upland mixedwood sites shared significant indicator species of
Cornus canadensis L.,
Geranium bicknellii Britt.,
Rosa acicularis Lindl.,
Linnaea borealis L., and
Elymus innovatus Beal with upland jack pine sites.
Vaccinium uliginosum L. and
Geocaulon lividum (Richards.) Fern. were unique indicator species in upland spruce communities.
The dominance of pre-fire and post-fire tree species (represented by log-ratios of compositions of basal area and stem density) were significantly different for jack pine, black spruce, and white spruce (Paired
t tests, Bonferroni-corrected *
p ≤ 0.04). When pre-fire dominance was characterized using total stem density (
Appendix A:
Figure A3), post-fire tree species compositions were significantly different for black spruce and trembling aspen (Paired
t tests, Bonferroni-corrected *
p ≤ 0.02). Having confirmed significant differences between pre-fire and post-fire tree species compositions, we examined the rank shifts in dominance of tree species in order to capture directionality of species-specific changes. Jack pine both increased and decreased in dominance in the post-fire cohort, but the slim majority of sites were neutral (−0.5 to 0.5 shift in rank; 39% of sites). Furthermore, not all plots burned with completely stand-initiating lethal wildfires (
Figure 3). Of the sites that experienced declines in the ranked dominance of jack pine, 41% had some surviving jack pine basal area post-fire (
Figure 3). Aspen dominance increased in the post-fire cohort, with 54% of sites gaining 1 or more ranks of dominance post-fire, and no sites declining by <−0.5 of a rank (
Figure 3). Both varieties of spruce primarily demonstrated no change or declines, in the post-fire cohort (57% of sites were neutral and 37% showed a decrease for black spruce; 69% of sites were neutral and 29% showed a decrease for white spruce). Of those sites with declines in the dominance of black spruce, only 9% (
n = 2) had incomplete mortality of black spruce trees (
Figure 3). No sites demonstrating declines in trembling aspen or white spruce had live individuals of these species post-fire (
Figure 3). Declines and increases in tree species dominance were significantly related to topoedaphic vegetation classes. Increases in jack pine dominance were especially associated with wetlands and spruce uplands, whereas jack pine declines occurred in upland communities where jack pine was already established, especially mixedwood stands where the suckering of trembling aspen was prevalent (
Appendix A:
Figure A4; Tukey test of least-squares means, *
p ≤ 0.05). Although increases in aspen dominance were more pronounced in uplands (Wilcoxon signed-rank test, **
p = 0.009), there were no statistical differences in aspen dominance shifts between topoedaphic vegetation classes (Tukey test of least-squares means;
Figure A4). Decreases in black spruce dominance were the most pronounced in upland spruce sites, whereas black spruce dominance was largely stable in treed wetlands and other vegetation classes (Tukey test of least-squares means, *
p ≤ 0.05). There were no significant differences in post-fire changes in white spruce dominance between topoedaphic vegetation classes (
Figure A4).
When representing the variance in understory vegetation and tree species dominance shifts explained by soils, we found that many soil properties were highly correlated. To address this, we decomposed the correlated soil physicochemical properties using PCA, and included only the first principal component (PC1) as an explanatory variable in the variance partitioning (
Table 2). We chose to retain percent sand and PC1, and organic soil depth and site moisture despite high correlations (
ρ = 0.8) between these two pairs, as these variables characterized important elements of the three environmental driver groups. Post-fire soils (Soils); pre-fire forests, topoedaphic context, and climate (Site); and burn severity and fire history (Fire) together explained 28% of the variance in understory vegetation communities, and 33% of the variance in the dominance shifts of tree species (environmental variables included in model reported in
Table 2;
Figure 4). There was a substantial shared variance explained between Soils, Site, and Fire. Overall, Site explained the largest portion of the variance in post-fire vegetation communities (8%) and tree species dominance shifts (13%;
Figure 4). Soils significantly explained 5% of the variance in understory vegetation but did not significantly explain tree species dominance shifts. Conversely, Fire was of substantial importance to tree species dominance shifts (7% of variance) but did not significantly explain post-fire vegetation communities (
p = 0.08;
Figure 4).
Classification and regression trees of the dominance shifts of tree species had
R2 values ranging from 0.65 to 0.43. Jack pine dominance increased in the post-fire cohort where the total soil N was ≥0.48% and decreased in the post-fire cohort in stands that experienced partial mortality (
Figure 5). These sites were typically mixedwood stands, and often had some remaining live basal area of jack pine trees, suggesting that these declines in the dominance of the post-fire cohort do not necessarily indicate persistent shifts away from jack pine dominance, although aspen suckering outpaced the establishment of pine seedlings (
Appendix A:
Figure A3). Trembling aspen increased in dominance in nearly all plots, but increases were somewhat limited in lightly burned plots and plots with higher N availability, both of which tend to be characteristic of wetlands (
Figure 5). Changes in black spruce dominance were neutral in young stands where black spruce was essentially absent pre-fire and in wetlands (TSO < 80.5). Declines in black spruce dominance were augmented in moderate aged (TSO < 103.5) uplands and stands experiencing severe canopy burning (CFSI ≥ 4;
Figure 5). White spruce dominance increased slightly in sites with low N availability (Total N < 0.11%) and lower moisture deficits (CMD < 191.5). White spruce dominance declined in historically drier sites, especially in those sites that experienced some canopy involvement in the fire (CFSI ≥ 2.65;
Figure 5).