The global distribution and drivers of wood density and their impact on forest carbon stocks

Abstract

The density of wood is a key indicator of the carbon investment strategies of trees, impacting productivity and carbon storage. Despite its importance, the global variation in wood density and its environmental controls remain poorly understood, preventing accurate predictions of global forest carbon stocks. Here we analyse information from 1.1 million forest inventory plots alongside wood density data from 10,703 tree species to create a spatially explicit understanding of the global wood density distribution and its drivers. Our findings reveal a pronounced latitudinal gradient, with wood in tropical forests being up to 30% denser than that in boreal forests. In both angiosperms and gymnosperms, hydrothermal conditions represented by annual mean temperature and soil moisture emerged as the primary factors influencing the variation in wood density globally. This indicates similar environmental filters and evolutionary adaptations among distinct plant groups, underscoring the essential role of abiotic factors in determining wood density in forest ecosystems. Additionally, our study highlights the prominent role of disturbance, such as human modification and fire risk, in influencing wood density at more local scales. Factoring in the spatial variation of wood density notably changes the estimates of forest carbon stocks, leading to differences of up to 21% within biomes. Therefore, our research contributes to a deeper understanding of terrestrial biomass distribution and how environmental changes and disturbances impact forest ecosystems.

Fig. 1. Observed wood densities across the global forest inventory plots and within gymnosperms, angiosperms, forest types and biomes.

a–c, Wood density distribution of gymnosperm (a) and angiosperm (c) species and influence of the proportion of angiosperms on CWD (b). The wood density distribution in gymnospermous species is narrower and has a smaller mean (~20% lower) than in angiospermous species. b, CWD increases with increasing proportion of angiospermous species in forest communities. We included 8,249 taxa with information on angiosperms and gymnosperms comprising 8,036 angiosperms and 213 gymnosperms, each with wood density information available at the species or genus level. d, Map of CWD observations for the ~1.1 million plots from the GFBi database. e,f, Box plots of observed CWD at the forest type (e) or biome level (f). Box plot shows the median, interquartile range and whiskers for data spread, excluding outliers.

Fig. 2. Phylogenetic tree and wood density information of global tree species.

The phylogenetic tree was constructed using the R package V.PhyloMaker, with wood density information available for 4,298 species (189 families from 55 orders). Wood density exhibits a strong phylogenetic signal (Pagel’s lambda = 0.92, P < 0.01, Blomberg’s K = 0.01, P < 0.01). The colours of the branches and the grey bars at the tips represent the wood density of each species. To identify orders that have significantly different wood densities compared to all other tree species, we conducted a two-tailed significance test by comparing the order-level wood density with 999 randomized wood density values from the entire phylogenetic tree. The coloured circle surrounding the phylogeny represents different orders. The filled blue/red circles inside the phylogeny indicate orders that show significantly (P < 0.05) lower (blue) or higher (red) wood densities relative to all the species. Numbers inside the circles represent the average wood density of the respective order.

Fig. 3. Global maps of wood density.

a,c,e, Wood density maps for all species (a), angiosperms-only (c) and gymnosperms-only (e). a, The community-level wood density map was derived from an ensemble approach, averaging the global predictions from the 200 best random-forest models. c,e, Angiosperm-only (c) and gymnosperm-only (e) wood density maps were derived from ensemble averaging of the global predictions from the 100 best random-forest models, respectively. b,d,f, Corresponding latitudinal trends in wood density aggregated for each 0.1 arc degree latitude: all species (b), angiosperms (d) and gymnosperms (f). Error ranges represent 1 s.d. either side of the mean. Maps are projected at 30 arcsec (~1 km²) resolution. Non-forested areas are displayed in grey. In the wood density maps for angiosperms (c) and gymnosperms (e), we correspondingly excluded pixels where angiosperms and gymnosperms constituted <5% of the entire community.

Fig. 4. Variable importance of the selected environmental metrics.

a–f, The environmental metrics are based on random-forest models (a,c,e) and linear partial regression models (b,d,f). a,b, Variable importance of the selected covariates across global forests, including angiosperm ratio to control for wood density differences between angiosperms and gymnosperms. c,d, Variable importance within angiosperm-only communities. e,f, Variable importance within gymnosperm-only communities. Mean decrease in accuracy values in a, c and e represents the relative contribution of each variable to CWD variation, whereby we averaged the values of 100 bootstrapped random-forest models. Bootstrapped partial regression coefficients for each variable (b,d,f) were calculated by averaging the partial regression coefficients from 100 multivariate linear models. All variables were standardized to allow for direct effect size comparison. In addition, we quantified the absolute effects of these covariates using partial regression analysis, as detailed in Supplementary Table 5.

Fig. 5. Comparison of global living tree biomass distribution using spatially explicit wood density data versus a universal wood density value.

a, The global distribution of living tree biomass (in tonnes per hectare), derived by integrating our wood density map with spatially explicit data on living tree volume, root mass fraction and biomass expansion factors. b, Percentage difference in estimated living tree biomass when comparing results derived using the global wood density map (from a) with estimates using a single, universal wood density value. The difference is calculated as the percentage change by subtracting the spatially explicit estimate from the universal estimate and then dividing by the spatially explicit estimate. Blue areas show regions where the universal estimate is higher, and red/orange areas indicate where the spatial estimate is higher. c, Percentage difference between the two biomass estimation methods across biomes. Box plots show the median, interquartile range and whiskers for data spread, excluding outliers.