Increasing atmospheric dryness reduces boreal forest tree growth

Abstract

Rising atmospheric vapour pressure deficit (VPD) associated with climate change affects boreal forest growth via stomatal closure and soil dryness. However, the relationship between VPD and forest growth depends on the climatic context. Here we assess Canadian boreal forest responses to VPD changes from 1951-2018 using a well-replicated tree-growth increment network with approximately 5,000 species-site combinations. Of the 3,559 successful growth models, we observed a relationship between growth and concurrent summer VPD in one-third of the species-site combinations, and between growth and prior summer VPD in almost half of those combinations. The relationship between previous year VPD and current year growth was almost exclusively negative, while current year VPD also tended to reduce growth. Tree species, age, annual temperature, and soil moisture primarily determined tree VPD responses. Younger trees and species like white spruce and Douglas fir exhibited higher VPD sensitivity, as did areas with high annual temperature and low soil moisture. Since 1951, summer VPD increases in Canada have paralleled tree growth decreases, particularly in spruce species. Accelerating atmospheric dryness in the decades ahead will impair carbon storage and societal-economic services.

Fig. 1. Description of tree-ring dataset.

a Distribution of sample sites. The background colour on the map at left illustrates the distribution of the aboveground biomass (AGB) across Canada’s forests, the distribution of sample sites for the tree-ring dataset encompassing the nine sampled species is represented by points. Canadian Ecoregions are illustrated on the map at right, with the hemi-boreal zone overlaid. Yearly temporal distributions of sampled (b) trees, (c) sites, and (d) species proportions.

Fig. 2. The relationship between annual growth fluctuations estimated from tree rings (BAI) and vapour pressure deficit (VPD).

a Means of t-values by ecoregions for VPD_t−1 (left) and VPD_t (right); and (b) proportions of non-significant, positive and negative t-values for VPD_t-1 and VPD_t, among convergent models, with the number N of corresponding sites. The density distribution of all t-values is illustrated in Fig. S2.

Fig. 3. Pointwise t-values of the regression between annual growth fluctuations estimated from tree rings (BAI) and vapour pressure deficit (VPD) for the species Picea mariana, Picea glauca, Pinus contorta and Picea englemanii.

Maps are displaying site-species t-values; a bidimensional interpolation was performed on a spatial resolution of 1 × 1 degree, using the inverse distance weighting method based on the 12 closest neighbours. Interpolations were bounded using boreal mask (all species map) and species distribution area (species maps)^,,.

Fig. 4. Output of random forest algorithm predicting t-values from the seven environmental, tree species and forest structure variables.

The x-axis retrieves the variable depth in the tree, and dot size represents variables occurrence as root node (i.e more frequent root node occurrence equals larger dot); the y-axis retrieves variables’ importance, measured as an increase in decision tree mean square error (MSE) when the variable is randomized. Results based on bootstrap with 500 decision trees (also see Table 1). The seven predictive variables are Species, Mean Age, Mean Basal Area (Mean BA), Summer Soil Moisture Index (Summer SMI), Mean Annual Temperature (MAT), Mean Annual Precipitation (MAP) and Elevation.

Fig. 5. Partial-dependence plots showing the marginal effects environmental, forest structure and tree species features have on the predicted sensitivity of growth to VPD across Canada’s boreal forest.

The partial dependence functions were developed using a random forest algorithm and illustrated whether the relationships between the target and the features are linear, monotonic or more complex. The thick-red lines illustrate the predicted t-value changes by the random forest model, which are influenced by alterations in the features displayed on the x-axis. The lower curves represent sample density, while the grey-shaded areas indicate 95% sample coverages. Species is considered as a categorical feature, with the partial-dependence function illustrated by red bars; the number of sampled trees per species is illustrated by the grey bars. The seven predictive variables are Species, Mean Age, Mean Basal Area (Mean BA), Summer Soil Moisture Index (Summer SMI), Mean Annual Temperature (MAT), Mean Annual Precipitation (MAP) and Elevation.

Fig. 6. Changes in the annual growth of Picea mariana, Picea glauca, Pinus contorta and Picea englemanii relative to summer VPD.

a Temporal trends in summer VPD (kPa / year) since 1951, measured daily temperature and precipitation using Kimball’s method. Percent growth change (in red) averaged across all sites relative to (b–d) prior- and (e) current-summer VPD (in blue). Shaded areas represent the 95% confidence interval. Sample sizes for panels (b–e) were respectively 9481, 4522, 2859, and 1632 trees. Annual growth changes are percent deviation from predicted values generated by the generalized additive mixed models representing the BAI variations unrelated to tree development stage (see Methods, Eq. 3). Correlative relationships (r) between growth changes and summer VPD; the 95% confidence intervals around r were computed from a bootstrapping technique that accounted for autocorrelation and trend in data, and regression slope (β₁) from ordinary-least-square regressions.