Increasing atmospheric dryness reduces boreal forest tree growth

Climate change will alter boreal forests capacity to store carbon, all the more so because they are dominated by cold-tolerant species and are warming faster than most other land areas (0.2 °C to 0.5 °C per decade over 1961–2015). Warming is expected to alter the net annual carbon uptake of boreal forests through changes to temperature-related variables (growing season length, drought severity, vapour pressure deficit, freeze frequency) that affect tree mortality, recruitment, physiology and growth. Across Canada's boreal forest, land carbon storage capacity varies due to local and regional differences in climate, vegetation, soils, surficial geology, and disturbance regime. The complexity of the boreal environment and the diversity of boreal species produce notable contrasts in productivity responses to warming. Across boreal Canada, forest productivity is anticipated to decrease in the west because of water stress, and increase northeastward where temperature is currently the main factor limiting tree growth. Such mechanisms have been raised as potential drivers of forest greening or browning in Canada. Warming can thus enhance plant growth in cool and wet environments, but impair it above a threshold of water stress. Additionally, differential trajectories among tree genera, species and ecoclines are possible, depending on their stress tolerance and plasticity. Atmospheric vapour pressure deficit (VPD), the difference between the amount of water vapour in the atmosphere and the potential amount held at saturation, is a major temperature-related determinant of plant physiology. Without a corresponding increase in the actual amount of atmospheric water vapour, VPD will increase with warming because the saturation vapour pressure is a curvilinear function of air temperature. High VPD can induce high leaf and soil water loss, increasing the risk of plant water deficits and stress. Water deficits increase xylem tension and tissue cavitation: these negative effects are minimized by stomatal closure, which decreases water loss at the cost of carbon uptake, leading to growth reductions and eventually tree mortality. From a hydrometeorological perspective, higher atmospheric dryness also increases atmospheric demand for water from the land surface, increasing evaporation and reducing available soil moisture. Hence, plant growth responses to atmospheric dryness can result from direct effects on stomatal conductance and indirect effects on soil moisture through increasing evapotranspiration. Tree species have evolved different strategies to cope with atmospheric dryness and therefore have differential responses to VPD changes, modulated by environmental conditions. There is thus a need for in-situ, multi-species comparisons to quantify these differences across a broad range of environmental conditions. Recent remote sensing studies have produced mixed results about the relative role of VPD on productivity and carbon uptake. Tree ring studies offer an important opportunity to understand VPD impacts, especially pertaining to aboveground carbon uptake. To date, tree-ring analyses directly linking changes in dryness with boreal forest growth have only focused on a few species and areas. Integrating tree-ring sampling into national forest inventories has enabled large-scale ecological assessments related to forest health and carbon cycling, including in Canada. Quantifying large-scale tree growth responses to VPD would improve our understanding of forest growth responses under climate change and reduce uncertainties in the model predictions used to evaluate strategies for mitigating and adapting to climate change. Here, we assess boreal forest responses to changes in atmospheric VPD using a well-replicated tree-ring network covering Canada's forests over the period 1951–2018. The growth-VPD relationships enabled mapping of spatially-explicit VPD responses across Canada's boreal zone. We then explored the main drivers of differential growth responses to VPD, including species, local precipitation and temperature, elevation, and tree age and size. Finally, we determined how VPD and growth are changing over time for the most responsive species.

Study design and area: The study covers eleven forested ecozones across Canada, spanning boreal and hemi-boreal regions with strong climatic gradients. Tree-ring data were obtained from 32,189 trees sampled across Canada, integrated from national forest inventories and related datasets.

Response variable and species: Annual basal area increment (BAI) was calculated from tree rings and used as the growth metric. Analyses considered nine widespread boreal/hemi-boreal species, including Picea mariana, Picea glauca, Picea engelmannii, Pinus banksiana, Pinus contorta, Pinus resinosa, Populus tremuloides, Abies lasiocarpa, and Pseudotsuga menziesii.

Climate and hydrometeorological data: Daily maximum/minimum temperature, precipitation, and relative humidity for 1950–2018 were obtained via BioSIM v10.3, which interpolates site-specific estimates from Environment and Climate Change Canada station data. Vapour pressure deficit (VPD, kPa) was computed following FAO-56 (Allen et al.) using daily temperature and dewpoint (dewpoint estimated per Kimball et al.). A soil moisture index (SMI; % of water holding capacity) was estimated using a quadratic-plus-linear water balance model (Régnière et al.), parameterized with critical and maximum available soil water of 300 and 400 mm. All daily variables were aggregated to summer (June–August) by year.

Spatial processing: Climate and SMI data were interpolated to a 1° × 1° grid (n = 1705 grid points) and linked to the nearest tree-ring sites. Long-term site climatologies (1951–2018) were computed, including mean annual temperature (MAT) and mean annual precipitation (MAP). The study notes the relatively sparse weather station network in Canada and the trade-offs of using gridded data.

Statistical modeling: For each species-site combination, Generalized Additive Mixed Models (GAMMs) were fit to relate annual BAI to summer VPD of the year of ring formation (VPD_t) and the prior year (VPD_t−1). Convergence was achieved in 3,559 of 4,931 models. Model goodness-of-fit was evaluated by r^2 comparing observed vs predicted year-to-year growth fluctuations. Significance and sign of VPD effects were summarized using t-values for VPD_t and VPD_t−1 terms. Partial models controlling for SMI effects (partial BAI–VPD GAMMs) were also assessed to test robustness of VPD effects independent of soil moisture co-variation.

Determinants of VPD sensitivity: A Random Forest regression (500 trees; bootstrap) predicted the significant t-values from seven predictors: species identity, site mean tree age, site mean basal area (BA), summer SMI, MAT, MAP, and elevation. Variable importance was quantified by mean increase in MSE upon permutation; variable depth and frequency as root node were also computed to assess predictor influence.

Mapping and visualization: Spatial patterns of VPD-growth responses were mapped using inverse distance weighting (IDW) interpolation of site-species t-values at 1° × 1° resolution, displayed overall and for key species, bounded by boreal and species distribution masks.

Data and code: Weather data are available from ECCC and BioSIM; tree-ring data from Natural Resources Canada’s TreeSource repository; analysis code is available on GitHub and Zenodo.

• Model performance and prevalence of VPD effects: 4,931 species-site GAMMs were fit; 3,559 (72%) converged with mean r^2 = 0.51 (σ ± 0.21, n = 3,559). Among converged models, 58% (2,057) showed a significant relationship between BAI and VPD_t, VPD_t−1, or both across nine species.
• Current-year VPD effects: 31% (1,096) of converged models had significant VPD_t effects (p < 0.05); 752 (~69%) were negative, concentrated near warm, dry margins of the boreal forest (southern Boreal Shield), especially for Picea glauca and Picea mariana. The remaining 344 significant VPD_t effects were positive, found mainly in western Canada (Boreal Cordillera) and cooler, moister regions (Hudson Plain, Taiga Shield).
• Prior-year VPD effects and lag: 47% (1,660) of converged models had significant VPD_t−1 effects; 96% were negative, indicating a strong lagged detrimental impact of elevated VPD on subsequent-year growth.
• Overall detrimental impact: VPD (current and/or prior year) had a significant and detrimental impact on BAI in approximately 51% of the 3,559 species-site combinations (~37% of the initial 4,931 models).
• Determinants of sensitivity: Species identity was the most influential predictor (lowest average depth; root node in 158/500 trees). More negative responses (higher sensitivity) were observed in order: Pinus resinosa, Pseudotsuga menziesii, Pinus contorta, Picea glauca, Picea mariana, Pinus banksiana; less negative in Populus tremuloides, Picea engelmannii, Abies lasiocarpa. Mean tree age was the next key determinant (root in 68/500), with younger to mature stands (0–100 years) showing more negative responses; sensitivity decreased with age, plateauing around 200 years. Among environmental variables, MAT was primary (root in 117/500; strong importance), with higher MAT associated with more negative t-values (i.e., stronger negative growth response). Lower summer SMI also aligned with more negative responses; MAP and elevation had lesser influence; mean BA contributed minimally.
• Robustness to soil moisture confounding: Significant VPD effects on growth persisted in partial models controlling for SMI, with ~35% of species-site models remaining significant, indicating VPD impacts are not solely artifacts of soil moisture co-variation.
• Spatial and temporal patterns: Negative VPD-growth relationships predominated in warm, dry southern boreal margins; positive relationships occurred in colder, wetter regions where warming-associated VPD increases may alleviate cold/excess water limitations. Since 1951, increases in summer VPD across Canada have paralleled decreases in tree growth, particularly in spruce species.
• Mechanistic interpretation: Results are consistent with direct stomatal closure under high VPD reducing carbon uptake and indirect effects via reduced soil moisture; a consistent lag suggests carbon allocation dynamics (e.g., NSC storage/use) and/or soil moisture memory effects.

The study demonstrates that increasing atmospheric dryness (higher VPD) has already imposed widespread constraints on boreal tree growth in Canada, addressing the central question of how VPD influences growth across species and environments. Both concurrent and lagged VPD effects were prevalent and predominantly negative, indicating immediate physiological constraints via stomatal closure and a carryover reduction in growth the following year, plausibly due to altered carbon allocation (e.g., NSC dynamics) and hydrological memory. Species identity and age structure strongly modulate sensitivity, with conifers such as Picea and Pinus (notably P. resinosa, P. contorta, and widespread spruces) showing greater negative responses, and younger stands exhibiting higher vulnerability. Environmental context further governs outcomes: higher MAT and lower soil moisture intensify negative VPD effects, consistent with more conservative stomatal behavior under warmer, drier conditions that reduces carbon gain. Conversely, in colder or wetter regions (e.g., eastern Boreal/Taiga Shields, Boreal Cordillera), modest VPD increases associated with warming can coincide with positive growth responses, likely where warmth relieves temperature limitations and elevated atmospheric demand enhances evapotranspiration on saturated soils, mitigating hypoxia. These spatially heterogeneous responses help explain observed greening/browning patterns in northern ecosystems and align with the notion that the role of atmospheric demand in ecosystem water and carbon fluxes is increasing. Importantly, the persistence of VPD effects after controlling for soil moisture indicates a primary stomatal response to VPD, not solely an artifact of drought co-variation. The age-dependent sensitivity implies that climate change may disproportionately affect regeneration and young cohorts, with implications for forest resilience, productivity, and carbon storage. Collectively, as VPD continues to rise with warming, negative impacts on boreal tree growth—and thereby on ecosystem carbon sequestration and socio-economic services—are expected to intensify, particularly in warmer, drier parts of the biome and for sensitive species.

This study provides a national-scale, multi-species, tree-ring based assessment showing that rising atmospheric VPD has already reduced growth across large portions of Canada’s boreal forests, with both immediate and lagged impacts. Sensitivity varies by species (strongest in several conifers including spruce and pine), stand age (higher in younger stands), and environment (more negative under higher MAT and lower soil moisture). Positive VPD-growth relationships are limited to colder, wetter regions where warming alleviates other constraints. These findings imply that continued warming-driven increases in VPD will diminish growth of common boreal species, reducing forest resilience, carbon storage potential, and provisioning of timber and ecosystem services. Management and policy should account for heightened VPD sensitivity—especially in young stands and sensitive species—by adapting silviculture, regeneration strategies, and species selection to projected warmer, drier climates. Future work should refine mechanistic understanding of lag effects (e.g., NSC dynamics), integrate higher-resolution soil and microclimate data, investigate trait-based drivers of interspecific sensitivity (including hydraulic strategies), and incorporate VPD-sensitive growth responses into Earth system and forest management models for improved projections.

• Climate data limitations: Canada’s sparse station network necessitated use of 1° × 1° gridded/interpolated data, which may not capture fine-scale or elevational gradients precisely.
• Soil moisture representation: The Soil Moisture Index (SMI) is a simplified water balance metric with fixed site water-holding parameters (300–400 mm), introducing uncertainty relative to real site-specific soil properties.
• Species and site coverage: Analyses focused on nine species; while extensive (32,189 trees), results may not generalize to all boreal taxa or unsampled regions.
• Model convergence and structure: 72% of candidate GAMMs converged; non-convergent cases were excluded. GAMM formulations capture statistical relationships but do not fully resolve underlying physiological processes.
• Age distribution: Fewer sites with very old mean ages (>250 years) limit inference on VPD sensitivity in the oldest stands.
• Potential confounding factors: Although partial models suggest VPD effects are not solely due to soil moisture co-variation, other co-varying climatic or disturbance factors (e.g., insects, fire legacies) could contribute at some sites.
• Tree-ring inference: BAI from tree rings reflects aboveground wood growth and may not capture whole-plant carbon allocation changes or root dynamics.