Introduction

Recent climate change projections indicate that annual mean temperatures could increase by 4–5 °C by the end of the twenty-first century across Canada’s boreal zone, if there is little progress on controlling global anthropogenic greenhouse gas emissions (Price et al. 2013). Such temperature increases will significantly impact the structure and function of the forest landscape (Gauthier et al. 2015a). Changes in temperature and precipitation regimes are expected to shift the optimal climate zones of many boreal tree species northwards by hundreds of kilometers (McKenney et al. 2007; Prasad et al. 2013), altering forest composition and diversity (Heyder et al. 2011; Reich et al. 2015) and the goods and services derived from these forests (Lovejoy and Hannah 2005; Price et al. 2013; Gauthier et al. 2015a).

In particular, the transition zone between the southern boreal and northern temperate forests may be particularly sensitive to changes in climate and could experience major changes in species composition (Heyder et al. 2011; Reich et al. 2015) resulting in shifts in the geographic distribution of trees and hence regional biomes (Beck et al. 2011). However, projecting such shifts is not trivial as there are many agents of change affecting forest landscapes. Firstly, changes in temperature and precipitation will have direct effects on individual tree species performance, e.g., growth, mortality, productivity, and recruitment (Luo and Chen 2013; Girardin et al. 2015), which lead to changes in inter-specific competition and succession (Luo and Chen 2015; Price et al. 2015; Reich et al. 2015). Moreover, climate-induced changes in regional natural disturbance regimes may trigger more rapid alterations in forest communities (e.g., Keane et al. 2013). An increase in forest fire activity (Boulanger et al. 2014) or shifts in pest outbreaks (Boulanger et al. 2016) will likely change the prevalence of young forests dominated by early successional species (e.g., aspen, birch and jack pine), and trigger changes in net ecosystem productivity (Girardin et al. 2011; Chen and Luo 2015), while also facilitating northward range expansion of temperate tree species able to take advantage of canopy openings (Matthews et al. 2011; Woodall et al. 2013).

As shifts in climate and their effects on disturbances are likely to be spatially and temporally heterogeneous throughout the Canadian southern boreal region (Price et al. 2013), one can expect varying responses of the forest vegetation (Bergeron et al. 2014). For instance, annual area burned could increase by more than 300 % to reach values of 2 % or more per year in central Canada, which is much higher than in eastern Canada (Boulanger et al. 2014). Central and western boreal regions are more likely to suffer from ecologically significant water deficits, causing serious impacts on drought-sensitive tree species (Hogg and Bernier 2005; Price et al. 2013; Wang et al. 2014; Ireson et al. 2015). Regional variations in the current distribution of vegetation, and in functional traits of the constituent species, e.g., dispersal ability, post-disturbance reproductive ability and climate-related productivity, could also shape a spatially heterogeneous response in the changes to forest landscapes (Matthews et al. 2011). Physiognomic (e.g., geographic barriers, soil characteristics, Gewehr et al. 2014) and anthropogenic factors (e.g., harvesting, fragmentation; Scheller and Mladenoff 2008), may also interact strongly with climate change-related impacts. Therefore, some regions will likely be at greater risk than others. As such, a broad holistic and spatially explicit assessment is essential when assessing the range of potential future changes in southern boreal forest landscapes. Such an assessment should be of great benefit to managers and policymakers for developing and implementing successful adaptations of present-day forest management practices.

Large-scale impacts of global climate change on forest ecosystems are commonly assessed using correlative species distribution models (SDM) (e.g., Iverson and Prasad 2002; McKenney et al. 2007). SDM models are simplistic but can provide useful ways to incorporate future conditions easily into conservation and management planning and decision-making (Pearson and Dawson 2003; Wiens et al. 2009). However, they have well-known limitations (e.g., Thuiller et al. 2008) including the inability to account for community-level interactions (Franklin et al. 2016), dispersal, and landscape-scale disturbances (but see Engler and Guisan 2009; Prasad et al. 2013). Many of these limitations may be overcome through the use of more mechanistic forest landscape models (FLM; e.g., Scheller and Mladenoff 2005). These models allow for succession to emerge from stand-level processes and landscape-level disturbances, which are key to projecting plausible impacts of climate change (Gustafson 2013). FLM may be linked to finer-scale patch models (e.g., Linkages, JABOWA) within which fine-scale ecological processes that integrate important tree responses to climate change (e.g., differential seed germination rates, tree–tree interactions, etc.) are modeled (Taylor and Chen 2009).

In this paper, we aimed to integrate a well-established gap-model (PICUS) and a well-known FLM (LANDIS-II) to explore and assess the impacts of natural and anthropogenic disturbances as well as climate-induced changes on forest landscapes along a longitudinal gradient across Canada’s southern boreal forest. Specifically, we assessed how these agents of change, when combined, will modify the total live aboveground biomass (AGB, t ha−1) and annual aboveground net primary productivity (ANPP, kg m−2 year−1 as well as forest communities from changes in species-level AGB (t ha−1). We hypothesized that the effects of a generally warmer and drier climate on growth, combined with sharp increases in disturbance frequency, will decrease total AGB and productivity in most regions. Concurrently, the increase in disturbances should promote increasing dominance of pioneer species over late-successional species and a warmer climate should favor warm-adapted mesophytic species at the expense of typical boreal species. We expected that western regions would be most at risk from these combined effects.

Methods

Study regions

Forest landscapes were simulated within four study regions located at the transitions between boreal and hemiboreal zones in each of four forested ecozones in Canada. These are, from west to east, the Boreal Plains (BP), the Boreal Shield West (BSW), the Boreal Shield East (BSE) and the Atlantic Maritimes (AM) (Fig. 1). Ecozones are the largest units of the National Ecological Framework of Canada (NEFC; Ecological Stratification Working Group 1996). The eastern and western parts of the Boreal Shield ecozone differ substantially in climate, however, and were split following Kurz et al. (2013) and Price et al. (2013). Each study region covers a transition zone between the southern edge of the boreal and the northern edge of the hemiboreal except for the AM which, because of its ocean proximity, lies entirely within the hemiboreal zone (Brandt et al. 2013). The area of each study region was set to span approximately 10 % of the “parent” ecozone. Each study region encompasses a wide variety of forest types, soils, and local climates, as well as anthropogenic and natural disturbance conditions (Supplementary Material S1 and S2).

Fig. 1
figure 1

Locations of the four regions where forest landscapes were simulated (delineated in red). There is one region for each of the following ecozones (delineated in black). AM Atlantic maritime, BSE boreal shield east, BSW boreal shield west, BP boreal plains. The boreal (light gray) and hemiboreal (dark gray) forest (sensu Brandt 2013) are also shown. (Color figure online)

Full size image

Climate data

Monthly time series of current climate were interpolated from climate station records using the data of McKenney et al. (2013). Future climate projections were obtained for the Canadian Earth System Model version 2 (CanESM2) using data downloaded from the World Climate Research Program (WCRP) Climate Model Intercomparison Project Phase 5 (CMIP5) archive for each of three different radiative forcing scenarios, known as Representative Concentration Pathways (RCP, e.g., van Vuuren et al. 2011), namely RCP 2.6, RCP 4.5 and RCP 8.5. The RCP 2.6 scenario represents a situation where radiative forcing peaks at ~3 W m−2 before 2100 and then declines to reach 2.6 W m−2 by 2100. In the RCP 4.5 scenario, radiative forcing is assumed to stabilize at 4.5 W m−2 after 2100 without an “overshoot” pathway. Conversely, in the RCP 8.5 scenario, the forcing reaches 8.5 W m−2 in 2100 and continues to increase for some time afterwards. According to these scenarios, mean annual temperature would increase by about 3.5 °C (RCP 2.6) to 7.5 °C (RCP 8.5) throughout the southern boreal region by 2100 (compared with circa 2000), while average precipitation is projected to increase by 10–25 % regionally with very small differences among forcing scenarios (Supplementary Material S2). Data from CanESM2 for the 1900–2100 period were bias-corrected by expressing them as differences from (temperature) or ratios of (precipitation) the CanESM2 monthly means for the 1961–1990 period.

The forest landscape model LANDIS-II

LANDIS-II is a spatially-explicit raster-based forest landscape model that simulates disturbances, seed dispersal, and forest succession (Scheller et al. 2007). Species are defined using unique life-history attributes and are represented in each grid cell as 10-year age-cohorts. Cell resolution was set to 250 m (6.25 ha). Forest composition and structure in each cell were initialized using forest properties data derived from the Canadian National Forest Inventory (NFI) and cohort data from provincial permanent and temporary forest inventory plots (FIP). Using species biomass as well as mean annual temperature and total annual precipitation as variables, we performed a nearest neighbour spectral analysis to attribute the FIP showing the smallest Euclidean distance to each 250 m cell. This imputation was conducted on a 20-year age class basis to ensure that the Euclidean distance between FIP and the 250 m cell was mostly due to site productivity and not stand age. Each of these cells was then assigned to a spatial unit (i.e., “landtype”) with soil (Mansuy et al. 2014) and climate conditions considered homogeneous. Grid cells with more than 50 % of their area covered with non-forest cover types were classified as inactive.

Forest succession and species growth potential

Forest succession in each grid cell was then simulated using the LANDIS-II Biomass Succession extension v 3.1 (Scheller and Mladenoff 2004). This extension takes into account tree species cohort age, life history traits, and species-specific landtype responses, and simulates changes in cohort biomass over time as each cohort regenerates, ages and dies. Species life-history traits information was collected using various sources (e.g., Burns and Honkala 1990 and previous LANDIS-II publications) including several previous studies involving LANDIS-II for North American forest landscapes (Table 1).

Table 1 LANDIS-II input data for tree species simulated within the four study regions
Full size table

To upscale physiological and demographic responses of tree species through time under different climate change scenarios, we parameterized our LANDIS-II simulation experiment with the individual tree-based, forest patch model PICUS, version 1.5 (Lexer and Hönninger 2001). PICUS simulates the dynamics of individual trees on 10 × 10 m patches across forest stand areas (generally 100–1000 patches, corresponding to total areas of 1–10 ha). It accounts for spatially explicit interactions among patches via a 3D light module, and simulates seed dispersal explicitly, as well as the effects of climate and soil properties on tree population dynamics (Lexer and Hönninger 2001; Seidl et al. 2005). Selected species-specific parameters used in PICUS may be found in Table 2. Complete description of the model and how it was parameterized and validated can be found in Supplementary Material S3). PICUS was used to derive three sets of species-, climate-, and landtype-specific dynamics inputs used in Biomass succession extension in LANDIS-II, i.e., (i) species establishment probabilities (SEP), (ii) maximum possible aboveground net primary productivity (maxANPP), and (iii) maximum AGB (maxAGB). Using individual tree information from the NFI and selected Canadian provincial inventory plots, we parameterized PICUS for 17 tree species occurring in the study regions (Table 1). To be included, these had to represent at least 0.3 % of total AGB of a given study region according to the 2001 NFI forest properties maps at 250 m resolution (Beaudoin et al. 2014).

Table 2 Select input parameters specific to PICUS for species simulated within the four study areas
Full size table

Using the PICUS model, we simulated mono-specific 1-ha stands for each of the leading tree species. A factorial simulation design was used to simulate all mono-specific stands for each study region, tree species and landtype under climate conditions for specific periods (2000–2010, 2011–2040, 2041–2070, 2071–2300) and forcing scenarios (baseline, RCP 2.6, RCP 4.5, RCP 8.5). All stands were simulated for 300 years from bare-ground and used the local soil (Mansuy et al. 2014) and climate time-series data. Values for SEP, maxANPP and maxAGB were then derived from these simulations (see Supplementary Material S4 for computational details).

Validation of dynamic growth parameters, as well as static growth- and mortality curve shape parameters, were assessed under baseline climate conditions by (i) assessing the realism of emerging successional pathways using pixel-level simulations as well as (ii) comparing species-specific biomasses at time t = 0 with the actual biomass estimated from the NFI forest cover maps (Beaudoin et al. 2014). Pixel-level simulations were conducted for one specific landtype chosen subjectively from the most common landtypes in a given study region to represent one general type of ecosystem (e.g., softwood boreal forest, mixedwood temperate forests, etc.). The successional pathways resulting from 1000-year simulations (Supplementary Material S5) showed good agreement with those reported in the literature (e.g., Racey et al. 1996; McLaughlan et al. 2010; Cyr 2014). The initial biomass for most species showed little discrepancy with NFI forest cover maps (Supplementary Material S6). However, balsam fir biomass was underestimated in AM, and the biomasses of jack pine and maples species were overestimated in most study regions. Despite these small discrepancies, we decided the overall calibration results were satisfactory given that the pixel-level simulations generally followed the known successional trajectories for these latter species.

Natural disturbances

Two natural disturbance agents, namely fire and spruce budworm (SBW, Choristoneura fumiferana [Clem.]) outbreaks, were considered in the LANDIS-II simulations. Together, these accounted for the majority of area disturbed by fire and insects in the study regions and both are recognized to have major impacts on Canada’s forest landscapes (e.g., Volney and Hirsch 2005). Fire simulations were conducted using the LANDIS-II Base Fire extension, which simulates stochastic fire events dependent upon fire ignition, initiation and spread. Fire regime data (annual area burned, fire occurrence, and mean fire size) were summarized into “fire regions” corresponding to the intersection of each region and the Canadian Homogeneous Fire Regime (HFR) zones of Boulanger et al. (2014). As we wished to constrain fire regime to known values for better traceability and reproducibility, fire simulations were not parameterized as a process emerging from dynamic changes in vegetation as well as from changes in climate. Baseline and future fire regime parameters within each fire region were calibrated according to models developed by Boulanger et al. (2014) and further updated for different RCP scenarios (Gauthier et al. 2015b, see also Supplementary Material S2).

Outbreaks of SBW were simulated using the Biological Disturbance Agent (BDA) extension v3.0 (Sturtevant et al. 2004), which is specifically designed to simulate host tree mortality following insect outbreaks. Host tree species for SBW included, from the most to least vulnerable, balsam fir (Abies balsamea), and white (Picea glauca), red (P. rubens) and black (P. mariana) spruce. Outbreaks are simulated as probabilistic events at the cell level with probabilities being a function of site and neighborhood resource dominance (e.g., host abundance within a 1-km radius) as well as regional outbreak status. Outbreak impacts (tree mortality) are contingent on these probabilities as well as on host species- and age-specific susceptibility. Parameters used in this study were calibrated and validated using various sources for the mixed boreal forest (e.g., MacLean 1980; Hennigar et al. 2008). Regional outbreaks were calibrated at highest severity level possible using this extension and were set to last at most one timestep (10 years) and to occur every 35 years in accordance with typical observed regional recurrence cycles (e.g., Boulanger et al. 2012).

Harvesting

Forest harvesting was simulated using the Biomass Harvest extension (v3.0; Gustafson et al. 2000). Historical harvest data (harvested AGB) were retrieved from MODIS-based annual (2001–2011) forest disturbance maps (Guindon et al. 2014) combined with the 2001 NFI forest properties maps, both at 250 m resolution to create estimates of cumulative AGB harvested during 2002–2011 at the 250-m grid cell scale. In this time period, grid cells where harvesting occurred along with estimated change proportion (%) were classified according to three levels of cumulative biomass removed (1–40, 40–80 and >80 %), hereafter considered as three harvesting prescriptions (Supplementary Material S1a). Mean harvested patch size, mean AGB harvested per prescription, and total harvested area, were summarized by “management areas” (i.e., either by forest management units for public lands or by ecodistricts for private lands). Older stands and those with the highest conifer proportion (according to species-level AGB) were given precedence for harvesting.

Simulation design

For each study region and climate forcing scenario, we ran five replicated LANDIS-II simulations for 300 years at 10-year time steps starting in the year 2000. SBW disturbance was not included in the BP region simulations because it is projected to remain a very minor disturbance factor in that region (Boulanger et al. 2016). Baseline parameters were used for the 2000–2010 period for all simulations as well as for the spin-up phase where the initial biomass of each species’ cohort is grown according to cohort presence-absence. Except for simulations forced by the baseline climate, only the climate-sensitive parameters (fire regime, maxANPP, maxAGB and SEP) were allowed to change in 2010, 2040, and 2070 according to the specific anthropogenic forcing scenario, but were held fixed thereafter up to the year 2300.

Analyses

For each region, we compared temporal trends in total and species-level AGB and ANPP, for each forcing scenario with those obtained for the baseline scenario. We also assessed the magnitude of regional-level changes in forest communities by calculating Bray-Curtis dissimilarity indices for species-level AGB data obtained from the baseline simulations compared to those from each of the future climate scenarios.

As frequently performed in similar studies involving LANDIS-II, total and species-level AGB, ANPP and community dissimilarity are reported here as mean values as stochastic variation among replicates was small. No formal statistical tests were performed; rather we assessed trends among areas and RCPs through visual inspection of graphs and comparisons of mean values reported.

Results

Projections of potential species growth using PICUS

Strong variations in growth pattern among time periods and forcing scenarios revealed important climatic constraints for individual species as simulated by PICUS, shown by strong variation in maxANPP, SEP and maxAGB (Supplementary Material S7). For several boreal species, climate-related potential growth was projected to decline with increasing forcing, but most markedly under the RCP 8.5 (Supplementary Material S7). The decline was particularly dramatic for balsam fir, black and white spruces and larch where growth was virtually nil for most landtypes under RCP 8.5 climatic conditions simulated for 2071–2100 (Supplementary Material S7). A few species (yellow birch, white pine, eastern hemlock, red and sugar maple, red oak and American beech) exhibited increased potential growth under most future climates, with the exception of RCP 8.5 for the 2071–2100 period, when potential growth was rather similar or even lower than estimated using baseline climate (Supplementary Material S7).

Projections of disturbance regime

Under baseline climate, annual area burned varied between 0.2 and 0.5 %, in the AM and BSW, respectively. As calibrated from the projections of Boulanger et al. (2014), annual area burned was projected to increase in all study regions, particularly under the RCP 8.5 climate scenario (240–610 % increase) and in the BSW where it reached 3.7 % year−1 by the end of the twenty-first century (Supplementary Material S8). Increases in area burned were substantially smaller under RCP 2.6 and RCP 4.5. The area affected by SBW outbreaks gradually decreased under all future climate forcings, but most markedly under RCP 8.5, most notably in the AM where it shifted from 35 % in 2020 to less than 5 % in 2300. Elsewhere, areas projected to be affected by SBM diminished almost to zero by 2300 under RCP 8.5. Much smaller declines in area affected by SBW were projected under RCP 2.6 and 4.5 in the BSE and BSW while they remained similar to those projected for the baseline climate in AM (Supplementary Material S8).

Total biomass harvested under RCP 2.6 and RCP 4.5 remained similar to that simulated under baseline climate (Suppl. Mat S8). However, for the RCP 8.5 climate scenario, total biomass harvested declined by 25–50 % relative to baseline by 2300, even though harvesting prescriptions were kept similar throughout the simulations.

Projections of ANPP and total AGB

A clear west-to-east gradient in overall changes in ANPP (Fig. 2) and total aboveground biomass (Fig. 3) occurred, with the largest decreases in both indicators projected for the western-most regions (BP and BSW). In all regions, projected decreases in ANPP (Fig. 2) and total aboveground biomass (Fig. 3) were inversely related to the radiative forcing. Compared with baseline, aboveground biomass and ANPP were projected to decrease by more than 80 % by 2300 in the BSW, and by 25–45 % elsewhere, under the RCP 8.5 forcing scenario, whereas both parameters remained unchanged or only slightly decreased under RCP 2.6 and RCP 4.5. Exceptions were for the AM where productivity was projected to increase by ~10 % between 2100 and 2150 under RCP 4.5. Changes relative to baseline occurred rather fast, with the largest changes occurring between ca. 2070 and ca. 2150 for most regions.

Fig. 2
figure 2

Trends in aboveground net primary productivity (ANPP) for each of the four regions simulated under baseline, RCP 2.6, RCP 4.5 or RCP 8.5 climate scenarios

Full size image
Fig. 3
figure 3

Trends in total aboveground biomass (AGB) for each of the four regions simulated under baseline, RCP 2.6, RCP 4.5 or RCP 8.5 climate scenarios

Full size image

Regional-level changes in forest composition

Projected changes in forest species composition relative to baseline climate were highly variable among forcing scenarios and study regions (Fig. 4), with the most dramatic (Bray-Curtis Index > 0.5) occurring in the BP and BSW under RCP 8.5. The eastern study regions appeared to be as vulnerable as the western regions to changes in composition under the RCP 2.6 and RCP 4.5 scenarios. For all regions, overall changes in species composition were largest under RCP 8.5.

Fig. 4
figure 4

Bray-Curtis dissimilarity indices for simulated forest communities projected under RCP 2.6, RCP 4.5 or RCP 8.5 climate scenarios compared with those simulated under baseline climate at the same timestep for each of the four study regions

Full size image

Regional trends in dissimilarities between communities experiencing baseline and future climate did not necessarily translate into similar regional changes in species-level biomass. Notably, changes in species communities in western regions (BP and BSW) were strongly driven by a shift in dominance to pioneer, disturbance-adapted species, especially under RCP 8.5 (Fig. 5). The net effect was to create generally younger forests composed of fewer species. In the eastern regions, however, the shift in forest composition resulted partially from an increase in the biomass of temperate, warm-adapted species (e.g., American beech, hemlock), and particularly of early to mid-successional species (e.g., red maple, white pine). These changes were rather small under the RCP 2.6 and RCP 4.5 but became more pronounced under RCP 8.5 (Fig. 5).

Fig. 5
figure 5

Stacked species aboveground biomass for each of the four regions simulated under baseline, RCP 2.6, RCP 4.5 and RCP 8.5 climate scenarios. See Table 1 for species abbreviations. Stacked species biomasses in the graphs are in the same order as in the legend

Full size image

Other important changes in forest communities were projected to be widespread in all regions. Generally, biomass of most boreal conifer species decreased relative to baseline under all RCP scenarios for all regions, although the decrease was particularly swift and large under RCP 8.5 (Fig. 5). These changes were rather abrupt, with larch, balsam fir, and black and white spruces projected to disappear under RCP 8.5 sometime after ca. 2100, except in BP where these species were projected to survive but with greatly reduced AGB (Fig. 5). Biomass of these species was projected to decrease less strongly for all study regions under the RCP 2.6 and RCP 4.5 scenarios. In general, forest composition was projected to become impoverished, leaving drought-tolerant jack pine and aspen as the dominant species. Red spruce and red pine also declined throughout all regions within their current range although the decline was much smaller under the RCP 2.6 and RCP 4.5 scenarios. For a few species (notably, sugar maple, yellow birch, balsam poplar, trembling aspen and white cedar), aboveground biomass remained rather similar relative to baseline under RCP 2.6 and RCP 4.5 although small to major decreases were projected under RCP 8.5 (Fig. 5).

Discussion and conclusions

To our knowledge, this is the first study to project the impact of climate change on several southern boreal forest landscapes by integrating a stand-level model, PICUS, with a spatially explicit forest landscape model, LANDIS-II, to improve the precision of projected ecological changes in Canada. Our study demonstrates that impacts of global warming are likely to prove very significant throughout the southern boreal transition zone, but they are subject to major differences between east and west.

Our models projected a swift and widespread decline of mid- to late-successional boreal species along the southern boreal transition zone with some abrupt changes projected for as soon as the next 50–100 years. Such a decline can be explained by both climate changed-induced decreases in potential growth and by rapid increases in area burned. Both modeling and field studies have shown that boreal species growing at the southern edge of their present-day distribution are likely to be vulnerable to warmer temperatures (Huang et al. 2013; Fisichelli et al. 2014). Along with consequential increases in water stress, warmer temperatures increase the occurrence of hot days which reduce tree’s photosynthetic capacity (e.g., Reich et al. 2015) while increasing metabolic respiration, hence inducing greater metabolic costs (Bond-Lamberty et al. 2014; Girardin et al. 2015). In this study, the PICUS model suggested that significant warming will impose strong constraints on the growth of several boreal species (notably balsam fir, white and black spruces and larch), sufficient to restrict survival by the end of the twenty-first century under RCP 8.5 (see Supplementary Material S7). Evidence for declining growth and increased tree mortality in recent decades within the southern boreal forest has been documented for black spruce (Girardin et al. 2015), jack pine (Luo and Chen 2015) and trembling aspen (e.g., Lapointe-Garant et al. 2010; Michaelian et al. 2011; Luo and Chen 2015). These declines were all partially linked to recent warming trends (but see also Price et al. 2015; Zhang et al. 2015). Important decline in growth and survival of boreal species has also been projected under severe radiative forcing scenarios using other modeling approaches (Girardin et al. 2015). Very short fire return intervals (<50 years) as projected for the RCP 8.5 radiative forcing, in at least parts of the BSE, BSW and BP, could also trigger the widespread decline of boreal late-successional fire-avoiders such as balsam fir (De Groot et al. 2003) and white spruce. Black spruce could also be affected by repeated fires, especially if the return interval becomes shorter than typical seed-bearing age (ca 30 years), curtailing post-fire regeneration (Brown and Johnstone 2012). Such a decline for these widespread species within the broad Canadian southern boreal transition zone could contribute to rapid modifications of forest communities over very large areas.

As a corollary, increased growth rates of co-occurring temperate species under most radiative forcing scenarios would also strongly reduce the competitive ability of boreal species at the temperate-boreal forest interface (Reich et al. 2015), favoring the northward expansion of warm-adapted species in Eastern Canada (i.e., in the AM and BSE). Many studies have already projected a general increase in temperate species in this transition zone (e.g., Scheller and Mladenoff 2008; Steenberg et al. 2013; Duveneck and Scheller 2015). Our simulations show the abundance of temperate species would increase primarily in eastern regions, where these species are already a significant component of the current species pool, where annual precipitation amounts are generally non-limiting, and northward colonization is more likely. In fact, recent warming may already have contributed to the recent expansion of these species in adjacent boreal patches (e.g., Boisvert-Marsh et al. 2014; Fisichelli et al. 2014). Although co-occurring temperate species might show a competitive advantage over boreal species in a warmer climate, excessive warming might impose further climatic constraints on temperate species even at their current northern cooler range limit as significant declines in growth potential were projected in our simulations for these species under RCP 8.5 beyond 2070.

Projected increases in biomass of pioneer and fire-adapted boreal species, especially under RCP 8.5, would result from projected increase in disturbed areas, particularly those burned throughout the southern boreal transition zone, and particularly in western regions. These species would comprise most of the surviving tree species in the BSW and BP where area burned is projected to increase by over 600 and 350 % under RCP 8.5, respectively (Supplementary Material S2). Furthermore, increased fire activity would likely enhance the northward expansion of some temperate species able to take advantage of canopy openings, including red maple (Zhang et al. 2014) and to a lesser extent white pine. Both species along with other temperate early-successional species were shown to have expanded their northward ranges recently (Boisvert-Marsh et al. 2014), notably by establishing more often in large gaps at the temperate-boreal transition zone (Leithead et al. 2010; Zhang et al. 2014). For instance, red maple regeneration at the northern edge of its range was shown to be highly contingent upon stand-replacing fires which would provide a favorable light environment (Zhang et al. 2014). These life-history traits evidently improve the capacity of temperate deciduous species to use opportunities for range expansion (Matthews et al. 2011). Increased fire activity could therefore be an indirect climate change trigger for invasion by temperate pioneer species by providing more suitable regeneration sites (Drobyshev et al. 2013; Boisvert-Marsh et al. 2014) and provide a significant competitive advantage (Reich et al. 2015) over boreal pioneers (e.g., jack pine, trembling aspen, and white birch) for which warmer climates impose strong constraints on potential growth (Supplementary Material S7).

The latter results emphasize the overwhelming impacts of disturbances, mainly fire, on future forest landscapes. Disturbances have been found to play a key role in defining how landscapes will respond to climate change (e.g., Keane et al. 2013; Seidl et al. 2014). Recent climate-induced shifts in fire regime mediated through changes in fire severity, fire seasonality and overall fire frequency have already contributed to changes in landscape patterns in Alaska (e.g., Kasischke and Turetsky 2006) and the southern Yukon (Hogg and Wien 2005). Other disturbances may yet play a minor role in such a shift. Indeed, the area affected by SBW outbreaks would actually decrease through the rapid climate- and fire-induced decline of main host species (e.g., balsam fir and white spruce). Harvested biomass would also decrease due to climate-induced declines in productivity. In this respect, changes in fire regime might represent one of the most important drivers in the successional dynamics of future boreal forest landscapes (De Groot et al. 2003; Keane et al. 2013).

Our projections strongly suggest that productivity and total biomass would decline rapidly, yet heterogeneously, along the southern boreal transition zone. One exception is for the AM region where total biomass would slightly increase from years 2000 to ca 2050 regardless of the climate scenario. The explanation for this could be that many of the landscapes in this region are relatively young as a result of past intensive harvesting (Supplementary Material S9). Otherwise, increased climate-induced growth potential for temperate species may barely compensate for the concomitant decrease of boreal species’ productivity at the regional level (Scheller and Mladenoff 2005; Corlett and Westcott 2013). The primary reason for this is that northward colonization by temperate species would occur at a much slower rate (Boisvert-Marsh et al. 2014) than the projected shifts in their climate envelopes (Iverson and Prasad 2002; McKenney et al. 2007). Such “migration lag” implies that markedly less productive forests will appear along the temperate-boreal transition zone, especially under RCP 8.5. Modified disturbance regimes would also be accountable for some decrease in productivity. In the present case, the increase in area burned would be so important that it would hasten the persistent establishment of many more young stands with low biomass and relatively poor regeneration, to a point that regional-level productivity would strongly decrease in the long term. In this context, significant increases in net carbon emissions from widespread climate- or fire-induced mortality should be expected (Kurz et al. 2013) in the short term. In the longer term, declines in forest productivity would greatly reduce the C sequestration capabilities of the southern boreal transition zone (Bond-Lamberty et al. 2007; Kurz et al. 2008).

Considering sharp changes in productivity and species communities, our results strongly suggest that western regions are at much greater risk than those located in the eastern part of the transition zone. Western regions have been shown to be highly sensitive to changes in climate and disturbance rates during the Holocene (Williams et al. 2009) and are projected to remain so in the upcoming decades with global warming (Hogg and Bernier 2005; Price et al. 2013; Ireson et al. 2015). Low-diversity forests, the scarcity of warmer-adapted species in present-day landscapes (Duveneck and Scheller 2015) and the potential for considerable climate-induced increases in fire activity suggest a strong decrease in forest productivity and density (De Groot et al. 2003; Frelich and Reich 2010), meaning a transition to forest conditions closer to those that prevailed during the Holocene (~4500–8000 cal year bp) in this area (Williams et al. 2009; Moos and Cumming 2011). Such a state shift from forest to parkland ecosystem can be expected to have profound impacts on ecosystem resilience, biodiversity and timber supply (Gauthier et al. 2015a).

Several limitations to this study are worth mentioning. Fire effects on forest communities are most probably overestimated as these were directly calibrated from HFR-based model projections (Boulanger et al. 2014) that do not take vegetation feedbacks into account. Such feedbacks are likely to be negative (Terrier et al. 2013), with recurrent fires promoting less flammable fuel (notably increased occurrence of young, predominantly deciduous stands). We did not include any impact of climate change per se on SBW population dynamics and hence on the vulnerability of SBW host species. SBW impacts are likely to be even lower than those simulated here as the climate envelope suitable for high SBW population growth rate would strongly shift northward with increasing climate forcing (Régnière et al. 2012). Also, possible CO2 fertilization effects on tree growth were not considered. The potential interaction of higher CO2 with significantly longer growing seasons adds uncertainty to predictions of reduced forest growth (e.g., Silva et al. 2010; Price et al. 2013; Arora et al. 2016). Given that productivity patterns were calibrated on 30-year averages of a future drier climate, our results may overestimate survival and forest productivity along the Prairie—boreal transition zone since potential widespread “pulses” of drought-induced dieback (e.g., Hogg and Bernier 2005; Peng et al. 2011) were not considered.

Changes in forest landscapes are likely to occur despite the fact that most boreal species have a rather high adaptive capacity through large environmental tolerance and population-level genetic diversity, and despite mechanisms (such as serotinous cones) evolved to cope with frequent disturbance (e.g., Aitken et al. 2008; Gauthier et al. 2015a). This demonstrates the high vulnerability of the southern edge of the boreal forest when facing a generally warmer and drier future regional climate. Anticipated consequences for forest goods and services would therefore be manifold and potentially severe (Gauthier et al. 2015a). The greatest uncertainty for the future of forest landscapes in Canada’s southern boreal transition zone results from the considerable variations among radiative forcing scenarios. This in turn reflects the dependence of future forest dynamics on global and national efforts to mitigate climate change. As such, we stress that conservation and forest management planning within the southern boreal transition zone must consider the impact of climate on both natural disturbances and growth-driven changes in vegetation communities: adaptation of these management practices will be crucial.