Wood traits related to size and life history of trees in a Panamanian rainforest

Sampling

A 50-ha forest plot was established in a primary tropical moist forest on Barro Colorado Island (BCI), Panama in 1982 and remeasured in 1985 and every 5 yr since (Condit et al., 2006). At BCI, the temperature averages 27.4°C, and annual rainfall averages c. 2600 mm with a four month dry season (< 100 mm rainfall per month). Because tree coring is prohibited on BCI, wood samples were collected nearby in similar tall (> 20 m), closed canopy, moist forests. In contrast to the primary forest on BCI, these were late secondary forests, mostly from the second half of the 19^th Century (and estimated to be 130–150 yr old), but a few stands from after the time of the Panama Canal construction (1914) or after WWII (1945). Generally, five mature trees per species were cored at breast height with standard 5-mm increment borers and the diameter at breast height (DBH, ranging from 6 to 207 cm) and the GPS coordinates were recorded. We use DBH_s to refer to the diameter of the cored tree and DBH_max for the maximum diameter of a species (see later). Cores were split into 5-cm segments and the wood density (WD, g cm⁻³) of all samples was determined by the water displacement method as mass after drying at 100°C divided by green volume (Wright et al., 2010; Hietz et al., 2013). We randomly choose one of the five cored trees per species for wood anatomical analysis (see later).

Leaves were sampled from the uppermost branches of the six tallest trees of each species in the 50-ha plot and used to measure leaf area (cm²), leaf mass per area (LMA; g m⁻²) and leaf nitrogen content (N_leaf; %) (Wright et al., 2010).

Anatomical analysis

We analysed the outermost segment – that is, wood within 5 cm from the cambium and mostly within 1 cm – from one individual for each of 325 tree species (species shown in Supporting Information Fig. S1). For statistical comparisons between wood anatomy and WD, we used the WD of the sample (WD_s) used for the anatomical measurements rather than mean density of the species. However, for analyses relating demographic traits and functional traits, we used the mean WD (WD_m) of the five sampled trees, with WD_s = 0.068 + 0.89 × WD_m, r² = 0.802. To calculate WD_m, each segment was weighted by the area of the annulus it represented, assuming a cylindrical trunk, so that WD_m reflects the density of a stem disk in which the small volume of wood in the centre contributes less than the larger volume of wood close to the bark.

We made 20–30-μm transverse sections with a sliding microtome, stained with methylene blue and embedded in Euparal (Carl Roth, Karlsruhe, Germany). Several photos were taken from each section with a microscope (DM4000M; Leica Microsystems, Wetzlar, Germany) equipped with a digital frame grabber. The lumen areas of vessels (hereafter called vessel area) were colour-coded manually using the ‘flood fill’ feature of Photoshop (CS3; Adobe Systems, San José, CA, USA). The area, maximum width and the width perpendicular to the maximum width of each vessel, as well as the total area of the section analysed, were then measured automatically with SigmaScan Pro (Systat Software Inc., Chicago, IL, USA). The area of the section analysed ranged between 1.7 and 140 mm² (mean 21) and included between 26 and 1362 (mean 240) vessels.

Data analysis

For each species, we calculated individual vessel area (VA, mm²); vessel density (VD, vessels per mm²) and F, which is the proportion of the section occupied by vessel lumina. Wood hydraulic conductivity (Kh, kg m MPa⁻¹ s⁻¹) was calculated according to the Hagen–Poiseuille law where Kh = ρw/(128 η) VD Dh⁴ 10⁶, where ρw is the density of water at 20°C (998.2 kg m⁻³) and η is the viscosity of water at 20°C (1.002 × 10⁻⁹ MPa s). Dh, the mean hydraulically weighted vessel diameter, was calculated as Dh = (Σ D⁴/n)^1/4, where D is the average of the major and minor axis for each vessel cross-section (in mm) and n is the total number of vessels measured (Tyree & Zimmermann, 2002). The calculated Kh based on the Hagen–Poiseuille law is substantially higher than true conductivity as it ignores resistance of water flowing through the vessel walls. However, vessel walls contribute a relatively constant 56% (± 2%) to total resistance in conduits (Sperry et al., 2006), thus calculated Kh appears to be a good proxy for true conductivity. Because vessels are generally elliptical and often circular, VA and D² are strongly correlated (r² = 0.99).

We analysed only one wood sample per species from a location different from the long-term monitoring plot. To assess the potential effect of within-species variation, we measured vessels in three individuals for 18 species to calculate within and among species variance, and used a Mantel test (Legendre & Fortin, 1989) to evaluate the potential effect of spatial variation (see Notes S1 for details).

Information on maximum tree size and demographic parameters were obtained from census data measured in the 50-ha plot on BCI in 1982, 1985, 1990, 1995, 2000 and 2005 (Wright et al., 2010). We calculated diameter growth (DGR) and mortality (M) rates for trees ≥ 100 mm DBH because wood samples were from the outer 5 cm of the cores and thus represent larger trees, and because DGR as well as wood structure can change rapidly in small trees. Individual-level DGR equals 100 × log_e(DBH_f/DBH_i)/∆t, where ∆t is the time in years between measurements. The subscripts f and i refer to final and initial values, respectively. Population-level mortality rates equal 100 × (1 − (N_f/N_i)), where N_i is the number of individuals present initially and N_f is the number of survivors (Wright et al., 2010). Relationships with growth rates and M were evaluated only for species with at least 50 observations. Maximum height (H_max) and maximum DBH (DBH_max) are the means of the six largest (by DBH) individuals of each species in the BCI 50-ha plot in 2007 (Wright et al., 2010).

Relative DGR is a useful measure to compare changes in size and growth trajectories between species, but not to assess differences in biomass investment because the same volume increment requires greater biomass for high-density than for low-density wood. To calculate biomass growth rates, we estimated the aboveground biomass (AGB) of trees using a general allometric model for tropical wet forest trees (Chave et al., 2014: AGB = exp(−1.803 − 0.976 × E + 0.976 × log(WD) + 2.673 × log(DBH) − 0.0299 × log(DBH)²) with DBH in cm and E a parameter depending on climate (E = 0.07072 for BCI). Starting with a DBH of 100 mm in year 0, DBH will be 100 × e ^DGR/100 in year 1. Using DGR for trees ≥ 100 mm DBH, total biomass growth rate (BGR) in 1 year is calculated as the difference in AGB between trees with diameters of 100 mm and 100 × e^DGR/100 mm.

We calculated ordinary least-square regressions to describe relationships between two variables, and used Fisher's r-to-z transformation to determine whether correlation coefficients (r) differed significantly between wood traits and DBH_max vs DBH_s.

We performed multiple regression analyses to evaluate the relationships between demographic traits (DGR, BGR and M) and wood and leaf traits (LMA, N_leaf, leaf area, Kh and mean WD_m). We performed a second multiple regression analysis to evaluate relationships between H_max and Kh, and mean WD_m. Leaves sampled from the uppermost branches of small species were shaded, those of large trees were sun-exposed and leaf traits vary with light intensities, and, for these reasons, leaf traits were not used to explain H_max. Because Kh was obtained from a single individual and wood traits are known to change with diameter, the starting model included an interaction term Kh × DBH_s. We used backward stepwise multiple regression analyses, eliminating insignificant variables from the model until only significant variables remained. The variance inflation factor, a measure of collinearity, was < 2 in the most complex models, indicating that collinearity between variables was not a problem. All variables except WD and H_max were log-transformed to meet assumptions of normality.

We calculated one principal component analysis (PCA) to evaluate associations among wood traits and a second PCA to evaluate associations among wood traits, leaf traits, DBH_s and demographic traits (DGR, BGR, M). The second PCA excludes maximum tree size because the pairwise regression analyses showed that DBH_s was a better predictor than DBH_max. Because vessel traits are strongly related, we used Kh as the only trait for vessels in the second PCA to prevent it from being skewed by an obvious trade-off between VA and VD. For the PCAs, data were mean-centred and scaled to have unit variance. Because the species for which wood samples were available were not all represented in the 50-ha plot and demographic parameters were used only for species with > 49 records, the second PCA was calculated with 98 species for which all variables were available.

We quantified phylogenetic signal for wood traits with Pagel's lambda (Pagel, 1999) using the ‘phylosig’ function in the R package ‘phytools’. We tested whether wood traits differed among 12 families with a minimum of ten species each with analysis of variance. We tested whether the slopes of the correlation between VA and VD differed among families and whether the regression slopes from our dataset differed from a large global dataset (Zanne et al., 2010) with standardized major axis (sma) tests using the R package ‘smatr’.

In order to test if correlations resulted from phylogenetic relatedness, we also calculated pairwise correlations using phylogenetically independent contrasts (PICs; Felsenstein, 1985). We constructed a phylogenetic tree using Phylomatic (Webb et al., 2008; http://www.phylodiversity.net/phylocom/) based on the APG3-derived megatree. We did not use the DNA-based phylogeny of 300 BCI trees (Kress et al., 2009) because this covers only about half of our species. We used estimated ages of nodes (Wikström et al., 2001) to date all other nodes by dividing branch lengths evenly between the dated nodes. Polytomies in the Phylomatic tree were resolved with the R function multi2di. PICs were calculated with the R package picante with PIC regressions forced through the origin. Statistics were calculated with R v.3.2.1 (R Development Core Team, 2011).

Relationships among wood traits

Wood anatomy and calculated Kh varied substantially among species (Table 1). Within-species variation of wood traits was about an order of magnitude lower than among-species variation. Spatial effects were insignificant for VA, DV and Kh (Mantel test, P > 0.9) and significant (P = 0.04) but low (r² = 0.0027) for WD (see Notes S1). This may be relevant because wood traits were not obtained from the site with the permanent plot, where life-history traits were obtained from.

WD_s was positively correlated with VD and negatively correlated with VA and Kh (Table 2). WD_s was not correlated with lumen fraction (F). Kh was strongly, positively correlated with VA (r² = 0.76), positively correlated with F (r² = 0.18), and negatively correlated with VD (r² = 0.31). F increased with VD (r² = 0.27), but was unrelated to VA (r² = 0.00) in spite of the very strong VD–VA relationship (r² = 0.77, Fig. 1). The slope of the relationship between VA and VD was −0.85 and significantly different from −1 (P < 10⁻⁸). The VA−VD slope did not differ between families in our dataset (P = 0.656) but did differ significantly (P < 10⁻⁵) between our dataset and a published global dataset (Fig. 1; Zanne et al., 2010).

The PCA with all wood traits showed most traits except F scaling predominantly on the first axis, with WD_s and VD scaling in one direction, and Kh, DBH_s, Dh and VA in the other direction. Only F scaled strictly on the second axis, orthogonally to VA, Dh and WD_s (Fig. 2).

Phylogenetic relationships

Phylogenetic signal (lambda) was high for VA (0.75), VD (0.79) and WD_m (0.77), but substantially lower for F (0.55) and Kh (0.50) even though both are derived from VA and VD. Family had a highly significant effect on all wood traits in one-way ANOVAs (P < 0.001).

PIC correlations were mostly similar to species-wise correlations (Table 2), which suggests that the relationships found were mostly not driven by phylogenetic relatedness.

Relationships with individual tree size and species-level maximum tree size

VA, VD and Kh were more strongly correlated with the DBH of the individual trees sampled than with the maximum DBH measured in the species (Fig. 3). For species-wise comparisons, the differences between correlation coefficients with DBH_max and DBH_s were significant (P < 0.05) for VA, Dh and Kh. For PICs, differences were significant (P < 0.01) for VA and Kh, and marginally significant (P = 0.07) for VD.

Relationship between functional and demographic traits

In multiple regression models, diameter and biomass growth rates were significantly related to WD_m, N_leaf and Kh, with Kh having the highest significance (Table 3). WD_m was negatively related to DGR but positively to BGR. LMA and leaf area were not significantly related to growth rates. WD_m was negatively related to M, and Kh and the interaction term (Kh × DBH_s) explained significant variation in H_max.

The first four axes in the PCA with wood, leaf and demographic traits as well as DBH_s explained 30.6%, 23.7%, 14.1% and 9.8% of the variation, respectively. WD_m, Kh and DBH_s had strong loadings on the first axis, and leaf traits had strong loadings on the second axis. DGR, BGR and M all scaled on the first and second axis, positively with Kh, DBH_s, LA and N_leaf, negatively with WD_m and nearly orthogonally with LMA (Fig. 4).

Do mechanical and hydraulic functions compete for available wood volume?

Although more vessels can transport more water, calculated hydraulic conductivity (Kh) and vessel density (VD) are negatively correlated because of the strong negative correlation between individual vessel area (VA) and VD (Fig. 1) and because Kh increases linearly with VD but with the square of VA. The negative correlation between sample wood density (WD_s) and Kh (r = −0.353, Table 2) remained highly significant when tree size was included, which suggests a trade-off between water transport and WD and hence a trade-off between water transport and mechanical support, which increases with WD (Niklas, 1992); however, the relationship between WD_s and vessel lumen fraction (F) (Table 2) was insignificant. Thus, water transport capacity does not appear to trade off with the cross-sectional area of wood available for mechanical strength. WD was also unrelated to vessel, fibre or parenchyma area for 42 tropical trees from Bolivia (Poorter et al., 2010).

By contrast, a strong negative correlation (r = −0.56) between WD_s and F has been reported for chaparral, a Californian vegetation with Mediterranean climate (Preston et al., 2006). Wood from chaparral might differ from rainforest wood because dry forest trees on average have higher wood densities than wet forest trees (Chave et al., 2009), which means that a larger proportion of wood volume is occupied by cell walls. At the same time, F was also greater for chaparral shrubs than for tropical humid forest trees. Allocation to mechanical strength (wood density) and to water transport capacity (F) may trade-off wood cross-sectional area in the chaparral plants with greater fractions of wood dedicated to both functions.

Why are WD and F unrelated in humid tropical forest trees? Cell walls contribute to wood biomass and thus WD, whereas cell lumina do not. WD must therefore be positively related to the fraction of cell walls and negatively to lumina. If vessel lumen fraction is not a driver of WD, WD must be driven by the lumen or wall fraction of other cell types. The main driver of WD appears to be fibre wall thickness (Jacobsen et al., 2007; Martínez-Cabrera et al., 2011) and a recent study found that the proportion of the cross-section occupied by all cell lumina or by fibre walls were the best predictors of WD (Fortunel et al., 2014). It appears that most trees can regulate WD and thus mechanical support without compromising vessel lumen fraction and hydraulic conductivity. Therefore, WD or lumen fraction are unlikely to constrain Kh. The correlation between WD and Kh is unlikely to reflect a direct trade-off but instead might be an indirect consequence of correlations of both with growth rates, where low WD enables fast growth and high water transport capacity supplies productive leaves.

Do species with large vessels dedicate a lower fraction of wood area to water transport?

We found a strong negative correlation between VA and VD similar to previous results from a large global dataset (Zanne et al., 2010); however, the slope and intercept of those relationships were significantly different (Fig. 1). At low VD, VA is almost 50% lower in our dataset than in the global dataset. We asked if this might be because our data from Panama represent exclusively tropical trees. For 409 tropical species in the global database (identified by selecting species classified as tropical in a global wood density dataset; Zanne et al., 2009), the slope of the VA−VD regression was marginally different from the slope from our dataset (P = 0.053), but regression intercepts differed significantly (P < 0.001) and VA at a given VD was still almost 50% lower in our dataset than in the global dataset.

It is difficult to assess the cause of this difference between datasets. We measured vessel lumen area and use mean diameter of individual vessels, whereas some studies in the global dataset report a range or the average of larger or smaller vessels. Moreover, the global dataset contains an unknown number of climbers, which often have large vessels (Carlquist, 1988) and a higher vessel density at a given vessel size (Jiménez-Castillo & Lusk, 2013). Including lianas in a dataset would thus result in disproportionally large vessels at the lower end of vessel densities, which is where our data differ most from the global dataset (Fig. 1). Datasets potentially obtained with different methods or from different growth forms should be compared with caution.

The implications for the vessel fraction (F) might be more interesting than possible differences in the absolute size of the vessels. In the data compiled by Zanne et al. (2010) with a slope of –1 for the regression between logVD and logVA, F does not change along the VA : VD axis. By contrast, we found an increase in F in Panamanian trees with smaller and more numerous vessels (Table 2), and a slope of the regression between logVA and logVD of −0.85, which is similar to that obtained from 42 species from a tropical rainforest in Bolivia (−0.87; data provided by L. Poorter). Because doubling vessel lumen area has a greater effect on Kh than doubling vessel density, species with few large vessels have higher Kh than species with many small vessels even though the total area dedicated to vessels is lower. Kh would be even larger in species with large vessels if F were constant over the range of vessel sizes and densities (Fig. 1). Apparently, species with larger vessels can reduce total lumen area.

Does phylogeny constrain relationships between wood traits and growth strategies?

Wood structure changes through evolution (Carlquist, 1975) and wood traits carry a strong phylogenetic signal. Based on a study of the California woody flora but not on statistical tests, it was suggested that vessel density evolves more rapidly than vessel area (Carlquist & Hoekman, 1985). We could not confirm this, as VD and VA both had similar phylogenetic signals. Interestingly, whereas phylogenetic signal was particularly high for structural traits (WD, VA and VD), it was substantially lower for Kh, which thus appears to be less constrained by evolution than vessel size and numbers. Hydraulic conductivity can be modulated by a combination of vessel size and area, and a small change in vessel area has a large effect on Kh. For this reason, Kh appears to be relatively unconstrained phylogenetically.

The relationship between wood traits and growth rates is also largely independent of phylogeny because the PIC regressions were mostly similar to species-wise regressions (Table 2). Likewise, whereas families differed strongly in individual traits such as VA, VD, F and Kh, the tight correlation between VA and VD also did not differ between families. Thus, whereas the variation in wood structure among species is large and constrained by phylogeny, the strategies of adjusting wood to enable fast growth are similar for unrelated species. Consequently, wood traits appear to be good predictors of growth strategies.

Does the relationship between tree size and wood structure invalidate correlations with life-history traits?

The relationship between wood structure and life-history traits might be affected by differences between the mature forest population on Barros Cerrado Island (BCI), where we obtained demographic traits from, and the late secondary forests off BCI, where wood samples were collected from. Many wood traits change with tree size and/or age (Lachenbruch et al., 2011), but whereas the wood samples were obtained from a younger forest than the one on BCI, the sample trees were mature trees and well within the size range of trees on BCI. In trees of a given size, wood traits can also differ if trees are growing under different conditions (Zobel & van Buijtenen, 1989), for instance growing isolated rather than in a closed forest. Because the late secondary forest was closed and of similar stature to the primary forest on BCI, and we used the outer wood –that is, the youngest part produced when the forest was already closed – we think this is unlikely to affect the relationship between the wood and life-history traits observed. Also, within-species variation was much lower than among-species variation and there was no or only a very small spatial effect on wood traits (see Notes S1), thus the sampling location appears unlikely to have affected the outcome of our study.

Correlations between Kh, F or VD with tree size were about twice as strong when using diameter at breast height for the sample rather than maximum for the species (DBH_s rather than DBH_max) (Fig. 4). Although the effect of tree size on wood structure is well known and can be studied readily via radial changes from the pith towards the cambium, tree size has often been ignored in interspecific comparisons of wood structure and relationships with other traits. Size-related changes occur because trees adjust wood structure according to the changing requirements imposed by the environment and ontogenetic stage (Lachenbruch et al., 2011). For instance, conduit lumen diameter tends to increase from juvenile wood to mature wood (Leal et al., 2007; Christensen-Dalsgaard et al., 2008; Lachenbruch et al., 2011). This ontogenetic increase compensates for the higher resistance imposed by a longer pathway in tall trees (Tyree & Zimmermann, 2002). This is often presented as a difference between juvenile and mature wood, but the change can be gradual and can take place over several decades (Tsuchiya & Furukawa, 2010 Schuldt et al., 2013).

If individual tree size is ignored, we may lose half the predictive power of functional traits (Fig. 4) or find and interpret differences among species that are in fact differences among trees of different sizes that disappear when tree size is included. Based on our finding that correlations between wood anatomical traits and tree size were much stronger for the size of the sample tree than for the maximum size achieved by the species, one may question previous interpretations of similar correlations (Poorter et al., 2010; Russo et al., 2010; Fan et al., 2012). Vessels and Kh reported from tree species that can attain a large size might be large because individuals sampled from these species tend to be large, and vessel size and Kh increase with tree size. Likewise, the widely held assumption that dryland plants have smaller vessels than plants from humid regions (Carlquist & Hoekman, 1985) might be the indirect result of associations between tree size and moisture availability, with smaller trees from drier regions, and between tree size and vessel diameters, with smaller plants having narrower vessels (Olson & Rosell, 2013). This calls for a re-thinking of the supposed benefits of small vessels.

Could the significant correlation between hydraulic conductivity and life-history traits also disappear when including tree size as a controlling variable? Although DBH_s is strongly correlated with Kh, the interaction between Kh and DBHs did not significantly influence the relationship beween Kh and growth rates (Table 3). The relationship between Kh and mortality was insignificant whether or not tree size was included. Only for maximum height of a species recorded in the 50-ha plot (H_max) did we find a significant DBH_s × Kh interaction, but H_max and Kh were still significantly correlated. The correlation between maximum tree size and Kh reported (Poorter et al., 2010) might result from sampling larger individuals in species that reach greater maximum size and the scaling of vessel size with tree size (Olson & Rosell, 2013). Still, efficient water transport is important for trees that grow tall as well as for species with high growth rates.

Trees also show strong ontogenetic changes in demographic traits, and functional traits predict differences in growth trajectories among species (Hérault et al., 2011; Rüger et al., 2012). It would be interesting to compare such ontogenetic changes in demographic traits with possible ontogenetic changes in functional traits. For instance, species with low WD had the greatest potential to modulate their growth (Hérault et al., 2011; Rüger et al., 2012), and these species also tend to show the strongest radial variation in WD (Hietz et al., 2013). Although the size of individual trees explained a large portion of the variation in wood traits (Fig. 3), information on ontogenetic changes in wood structure within a species is scarce. As with WD, species are likely to differ in their radial adjustments of Kh and other wood traits. An analysis of ontogenetic changes of wood traits for a wide range of species would help to better understand the functional significance of the traits as well as the life histories of the trees. Because wood structure is conserved (unless a tree rots), analysing radial changes in vessels or other components from the pith to the bark provides an easy pathway to study such ontogenetic adjustments.

Which traits best explain life histories?

We found that mortality rate (M), diameter growth rate (DGR) and maximum tree size all scaled negatively with WD (Table 2; Fig. 3), which agrees with previous studies (Wright et al., 2010; Martínez-Cabrera et al., 2011; but see Fan et al., 2012). Low WD enables rapid DGR because less biomass is required per volume grown. Surprisingly, for the question which trees efficiently produce biomass, we found that wood density was positively correlated with biomass growth rate (BGR), although the explanatory power of the multiple regression model for BGR was low (Table 3). This is unexpected because species with lower WD are often light-demanding with higher rates of photosynthesis (Santiago et al., 2004; Meinzer et al., 2008a). A positive relationship between WD and biomass growth was also found on relatively fertile soils in the Peruvian Amazon (Keeling et al., 2008). Keeling et al. offer two possible explanations for this: either species with high WD have deeper crowns and longer leaf lifespan, so that the overall carbon gain of their leaves may exceed that of lower WD species; or species with lower WD have higher rates of carbon loss through greater turnover of leaves or roots or through higher respiration rates. We also note that the relationship between WD and photosynthesis may be weaker than previously believed. This possibility is supported by the relative independence of wood and leaf traits including no significant relationship between WD and leaf nitrogen (N) (Baraloto et al., 2010). Unfortunately, many studies on tree growth evaluate only DGR. Calculating BGR would provide additional insight into the functional significance of traits.

WD was the best predictor of M. This may have multiple functional explanations, which are hard to tease apart in ecological studies because WD is related to mechanical strength, fungal resistance and cavitation resistance. At the study site in Panama, trees with low WD more commonly snapped whereas trees with high WD were uprooted (Putz et al., 1983), and in Amazonia species with lower wood density had higher drought-related mortality (Phillips et al., 2010). Several mechanisms are likely to be responsible for the relatively strong relationship between WD and M.

Kh is known to be related to diameter growth, but also varies with tree size, which may have confounded previous analyses (Poorter et al., 2010; Fan et al., 2012). Both studies used pairwise correlations including only wood traits to predict growth. Because Kh is affected by the size of the individual tree, our multiple regression models initially included a Kh × DBH_s interaction term but this was significant only for H_max. Apart from WD, the only leaf or wood traits with a significant effect on diameter as well as biomass growth in multiple models were leaf N concentration and Kh. Leaf N is strongly related to photosynthesis (Wright et al., 2004) and high water transport rates are necessary to supply productive leaves. Of all variables tested, Kh had the strongest relationship with growth rates, possibly because hydraulic conductance integrates transpiration per leaf area and total leaf area. For instance, Kh per sapwood as well as per leaf area is correlated with maximum photosynthesis, stomatal conductance, photosynthetic nutrient-use efficiencies as well as growth rates in 17 dipterocarp trees (Zhang & Cao, 2009). Plants generally show a high coordination between hydraulic and photosynthetic capacities (Brodribb, 2009).

We note, however, that Kh per sapwood area is not the same as water consumption per tree. The capacity of a stem to transport water equals the Kh per sapwood area times the total sapwood area. Interestingly, individual tree absolute biomass growth was best explained by sapwood area, which is interpreted as a positive effect of stem water storage (van der Sande et al., 2015), but could also result from a correlation between hydraulic conductance and sapwood area. Comparisons with tree water use would help to elucidate the causal relationship between Kh and growth.

Conclusions

Wood structure carries a strong phylogenetic signal, which means that the capacity for adaptations in the evolutionary short term is limited. However, wood traits evolve in a coordinated way and the fact that small changes in vessel size result in large changes in Kh means that trees are rather flexible in adjusting their hydraulic capacity. This is important during ontogeny, because tall trees can transport water more efficiently, and during evolution, because the water supply can be adjusted to the requirements of different growth rates. Of all wood and leaf traits, Kh was the best predictor of growth rates. However, because Kh also scales with tree size, the size of the sampled tree should be included in analyses of wood functional traits.