Northern expansion is not compensating for southern declines in North American boreal forests

Boreal forests store vast amounts of carbon and are a critical component of the climate system, yet boreal regions are warming at roughly twice the global average. Warming affects growth, mortality, recruitment, and disturbance regimes (droughts, wildfires, insect outbreaks), potentially undermining forest resilience and altering the biome’s distribution. Theory and observations suggest the boreal biome should shift northward: improved growing conditions and longer seasons could enable northern expansion into current tundra, whereas warmer and drier conditions at the southern margin may reduce growth and increase mortality of cold-adapted conifers, favoring temperate species and steppe-grassland encroachment. However, the rates of northern expansion (limited by dispersal and recruitment) may lag behind the potentially faster southern retreat (driven by stress and disturbances), creating a transient disequilibrium and possible biome contraction. This study uses satellite-derived tree cover to test three hypotheses across the North American boreal biome (Canada and Alaska) during 2000–2019: (1) tree cover changes are asymmetric, with slow gains at the northern edge and faster losses at the southern edge, leading to range shrinkage; (2) disturbance regimes (wildfire, timber harvest) and vegetation types modulate these changes, with stronger disturbance-related losses in the south and greater vulnerability of needleleaf forests; and (3) tree cover changes relate to temperature and precipitation gradients and their trends, with gains in colder, moister regions and losses in warmer, drier regions.

Prior work using spectral vegetation indices documented widespread greening in northern boreal regions and browning in southern areas, consistent with a northward shift, but such indices conflate canopy and understory signals and can suffer from cross-sensor calibration issues. Studies have observed woody encroachment into Arctic tundra and highlighted constraints on treeline advance (dispersal limitation, microsite suitability, facilitation by shrubs). Research indicates increased disturbance frequency and severity (wildfire, drought, insects) under warming, with potential to reduce resilience and accelerate southern declines. Growth responses to warming are heterogeneous: improvements at higher latitudes/elevations, but growth reductions and elevated mortality risks at warm, dry margins. Models often emphasize long-term equilibria and may miss transient dynamics such as lagged northern expansion vs faster southern retreat, implying potential biome contraction over human-relevant timescales.

Study area and sampling: North American boreal forests (Canada and Alaska). Sampling used 69 randomly placed south–north transects, each crossing both southernmost and northernmost boreal biome boundaries and extending 120 km beyond them. Sample plots were 0.05° × 0.05°. A minimum spacing of 133 km between transects was imposed based on an exponential spatial variogram of MODIS tree cover change to reduce spatial autocorrelation. Boreal biome boundaries followed Gauthier et al. Plots dominated by urban, cropland, wetlands, bare ground, water, or snow/ice were excluded; remaining dominant vegetation types were grouped as non-woody, shrubs, needleleaf, mixed, broadleaf, and unknown forest. Final dataset: 12,954 plots (~26.3 million ha). Tree cover and change: Annual tree cover (2000–2019) from MODIS Vegetation Continuous Fields v6 at 250 m resolution. For each plot, absolute tree cover change (% per year) was computed by Theil–Sen slope with Yue–Pilon pre-whitening to address temporal autocorrelation. Relative change was also computed by dividing absolute change by mean tree cover (2000–2019). Both absolute and relative changes were analyzed; results emphasized absolute change due to similar patterns. Boundary standardization: For each transect, the southern and northern intersections with the boreal boundary were assigned −1 and +1, respectively, with positions between them scaled accordingly; values beyond boundaries extended to −1.5 and +1.5. This standardized boundary distance (SBD) enabled consistent comparisons across transects irrespective of latitude or length. Disturbances and vegetation: Disturbance data (30 m) included Canadian wildfire and timber harvest (CanLaD) and Alaskan wildfire (MTBS), 1985–2019. For each plot, the percent area disturbed was computed, and plots were classified by dominant disturbance type and timing: (1) wildfire 2000–2019, (2) harvest 2000–2019, (3) wildfire 1985–1999, (4) harvest 1985–1999, or (5) undisturbed/older than 1985. Vegetation classes came from Copernicus Global Land Cover 2015 (100 m), reclassified to six main vegetation types. Environmental covariates: Climate from ERA5 monthly averages (0.25°), 1980–2019; annual mean temperature (MAT) and precipitation (MAP) and their trends for 1980–2019 and 2000–2019 were derived (Theil–Sen slopes). Elevation from USGS GMTED2010 (250 m). Statistical analyses: Spatial autocorrelation assessed via variogram; sampling ensured independence among transects. Relationships between tree cover change and SBD, climate, tree cover, precipitation, elevation, and trends were modeled using generalized additive mixed-effects models (GAMM; mgcv). Due to multicollinearity, predictors were modeled separately, each including interactions with disturbance category. Transects were random effects; an exponential spatial correlation structure accounted for within-transect autocorrelation. Additional GAMMs assessed interactions between SBD and the 30 combinations of vegetation type × disturbance category. Ecozone-level summaries used the Canadian Ecological Framework map.

• Biome-wide asymmetry: Tree cover increased overall but with strong spatial contrasts. Mean change across the boreal biome was +0.12% ± 0.40% SD per year (≈13,000 plots). Losses dominated at the southern boundary (−0.13% ± 0.39% per year), while gains increased towards the north and peaked in the northern half of the boreal interior (+0.22% ± 0.4% per year). At and beyond the northern boundary, changes were near zero (≈0% ± 0.07% per year).
• Range shrinkage: Regional changes produced a range shrinkage of tree cover distributions and a reduction in treed area around biome boundaries, indicating a declining latitudinal distribution range driven by southern losses despite densification in the northern interior.
• Ecozones: Southern/transition ecozones lost tree cover or had ~0 net change (e.g., Montane Cordillera −0.1% ± 0.39%; Boreal Shield 0% ± 0.5%; Boreal Plains +0.02% ± 0.43%), whereas northern/interior/higher elevation ecozones gained (e.g., Taiga Plains +0.27% ± 0.39%; Boreal Cordillera +0.26% ± 0.37%). Beyond the boreal boundary (Southern Arctic) ~0% ± 0.09%. Alaskan boreal forests: +0.08% ± 0.37% per year.
• Disturbances: In the absence of fire/harvest since 1985, tree cover increased across most of the range (+0.16% ± 0.24% per year), lower in the south (+0.04% ± 0.28%) and higher in the northern interior (+0.24% ± 0.25%). Where wildfire or harvest occurred since 2000, tree cover generally decreased; losses were larger for harvest (−0.15% ± 0.43% per year) than wildfire (−0.05% ± 0.48% per year). At the temperate–boreal transition, harvest drove larger losses (−0.3% ± 0.42%) than wildfire (−0.12% ± 0.33%); in the southern interior, wildfire caused −0.13% ± 0.5% vs harvest −0.06% ± 0.36%. Areas disturbed in 1985–1999 showed subsequent gains, especially after wildfire (+0.4% ± 0.35% per year) versus harvest (+0.16% ± 0.41% per year). At the transition zone, plots showed small losses after logging (−0.04% ± 0.34%) and minor losses after wildfire (−0.01% ± 0.24%).
• Vegetation types: Asymmetric south–north patterns persisted across non-woody, shrub, needleleaf, mixed, and broadleaf contexts. Needleleaf forests gained most in the northern boreal; areas dominated by non-woody or shrubs peaked in the interior. Mixed/broadleaf forests showed approximately linear gradients: gains northward, losses southward.
• Environmental gradients: Gains were largest at moderate mean tree cover (20–40%), intermediate MAT (−7 to 0 °C), and low–moderate MAP (~400–1000 mm), corresponding to interior/northern interior regions. Very cold/warm or very dry/wet areas and very open/dense forests saw losses or minimal change. Undisturbed areas with moderate warming showed notable gains. Precipitation trends had no clear relationship with change; elevation had little effect on absolute change.
• Interpretation: Slow gains at the tundra boundary and faster losses at the southern boundary indicate a mismatch between potential and realized distributions, consistent with a transient biome contraction onset.

The pronounced north–south asymmetry in tree cover dynamics suggests that northern expansion has not compensated for southern declines during 2000–2019, implying a disequilibrium between climate-driven niche shifts and realized forest distribution. Tree cover densification in the northern interior co-occurred with range shrinkage at biome edges, a hallmark of transient contraction. Disturbances (wildfire, logging) were key correlates of losses, especially in the southern boreal where recovery appears slower, consistent with warming-induced growth constraints and elevated mortality risk. In contrast, northern interior regions gained cover despite frequent fires, likely due to faster post-disturbance recovery and/or compensatory gains in unburned portions of plots and potential shifts toward fast-growing deciduous species. Climatic associations align with these patterns: gains in cooler, moderately moist interiors, and losses in warmer southern margins even without recent disturbance, indicating deteriorating growing conditions at the warm edge. At the cold northern boundary, expansion remains slow, likely limited by dispersal, recruitment, microsite availability, permafrost-related hydrologic changes, nutrient constraints, and herbivory, reinforcing the lag behind rapid climate change. The tree cover changes imply short-term biomass carbon uptake in the interior but losses at the southern margin; albedo decreases in the north and increases in the south may exert opposing climatic feedbacks. Overall, the dynamics highlight potential for a long transient contraction, with trajectory dependent on future warming, disturbance regimes, and the pace of northern expansion vs southern retreat.

Over the past two decades, North American boreal forests exhibited slow tree cover gains near the tundra boundary and rapid losses near the temperate boundary, leading to range shrinkage of tree cover distributions. Disturbances and temperature gradients were strongly associated with these changes. The lack of northern expansion to compensate southern declines indicates the potential onset of a transient biome contraction with implications for carbon storage, albedo, and climate feedbacks. Given the likelihood of continued warming and disturbance intensification, compensatory northern expansion may not occur on human timescales. Long-term, spatially comprehensive monitoring integrating remote sensing with field data, and improved models that capture transient dynamics, are essential to project biome trajectories, recovery capacities, and impacts on carbon and climate. Future research should: (1) extend multi-decadal observations of tree cover and structure with higher-resolution sensors and field validation; (2) integrate additional disturbance agents (e.g., insects) and their interactions; (3) quantify carbon and albedo feedbacks under contraction scenarios; and (4) assess management options to bolster resilience and recovery, particularly at the southern margin.

• Remote sensing constraints: MODIS VCF tree cover may miss low-stature trees/shrubs near the forest–tundra boundary due to height thresholds; mixed-pixel effects and resolution limits can obscure fine-scale dynamics. Cross-sensor validation is needed.
• Short temporal window: The 20-year period may be insufficient to fully capture long transients and slower demographic processes (recruitment, mortality lags) or multi-disturbance legacies.
• Disturbance coverage: Analyses of harvest data are limited to Canada; other disturbance agents (e.g., insects) were not comprehensively mapped at biome scales.
• Attribution: Observational design limits causal inference among climate, disturbance, and vegetation responses; multicollinearity among environmental variables required separate models for each predictor.
• Boundary/land-cover definitions: Static land cover map (2015) may not capture temporary or recent changes; boreal boundary delineation includes sparse woodland/tundra mosaics, potentially complicating interpretation.