1.Because of its mechanical properties, wood density may affect the way that trees expand their stem and crown to exploit favourable light conditions in a mechanically stable way. From engineering theory and wood density properties, it is predicted that in terms of biomass investment, low-density wood is more efficient for vertical stem expansion, while high-density wood is more efficient for horizontal branch expansion. So far, these predictions have rarely been tested by empirical studies.
2.We tested these predictions for 145 co-occurring tree species in a Malaysian tropical rainforest. For each species, we selected trees across a broad size range and measured architectural dimensions (stem diameter, height of the lowest foliage and crown width). We used a hierarchical Bayesian model to estimate species-specific allometric relationships between architectural dimensions including estimated stem biomass. Then, we examined correlations between species wood density and estimated architectural variables at standardized heights.
3.When species were compared at standardized tree heights, wood density correlated negatively with stem diameter and positively with stem biomass at most reference heights. This indicates that species with low wood density produce thicker stems but at lower biomass costs. Wood density correlated positively with crown width and negatively with height of the lowest foliage, which indicates that high wood density species have wider and deeper crowns than low wood density species. These relationships were maintained at most reference heights. However, the relationship with crown width was nonsignificant above 18 m height. This may reflect large plastic response of lateral crown expansion to a local condition.
4. Wood density explains the trade-off between effective vertical stem expansion and horizontal crown expansion across co-occurring tropical tree species. Such mechanical constraints characterize the difference in tree architecture between low wood density species that show an efficient height expansion to attain better light conditions in the exposed canopy and high wood density species that show an efficient horizontal crown expansion to enhance current light interception and persistence in the shaded forest understorey. Our study thus suggests that the mechanical constraints set by wood density contribute to the co-existence of species differing in architecture and light capture strategy.
A major challenge in community ecology is to understand how trade-offs between plant traits contribute to the maintenance of community structure and species diversity (Tilman 1982; Kohyama 1993; McGill et al. 2006; Muller-Landau 2010). In closed canopy vegetation with strong vertical gradients in irradiance, plants compete for better light conditions. Therefore, trade-offs between vertical and horizontal space exploitation should play an important role in the architectural differentiation among co-occurring species (Kohyama 1987; Hirose & Werger 1995; Falster & Westoby 2003; Poorter et al. 2003). Plants improve their fitness by maximizing the lifetime net carbon gain and successful offspring production (Givnish 1988; Sterck & Schieving 2007), which leads them to invest resources in durable structures to defend themselves against pathogens, to bear their own weight and to resist dynamic loading by wind and falling debris (Turner 2001). Tree species may do so by investing in dense wood. While denser wood allows trees to produce more slender stems keeping stable, it involves larger biomass investment per unit of height gain (Anten & Schieving 2010). A slender stem, however, reduces the amount of bark surface area, which may lower the cost of maintenance respiration at a given height, and thus increases the ability to survive at the low irradiance levels that are typical of the forest understorey (Larjavaara & Muller-Landau 2010). Therefore, stem wood density, defined as the dry mass per unit green volume of wood, may underlie the interspecific trade-off between effective height gain and a persistent life in the understorey (Kohyama 1987; Kohyama & Hotta 1990).
Species vary over one order of magnitude in their wood density, ranging from 0·08 to 1·39 g cm−3, and the encountered range of wood density is particularly large in species-rich tropical rainforests (Chave et al. 2006, 2009; van Gelder, Poorter & Sterck 2006). Wood density is a key functional trait, because of its importance for mechanical stability (King et al. 2006b), defence against herbivores (Loehle 1988), hydraulic conductivity (Meinzer 2003), photosynthetic carbon gain (Santiago et al. 2004) and diameter growth rates of plants (Poorter et al. 2008). From a mechanical perspective, wood density is strongly correlated with stiffness (i.e. Young’s modulus of elasticity) and resistance against breaking (Niklas 1992; van Gelder, Poorter & Sterck 2006). Because wood density is negatively related to volumetric growth and positively to strength, it is often assumed that there is a trade-off between height gain (with low construction costs by low-density wood) and mechanical stability (with high strength by high-density wood) (e.g. Poorter et al. 2003). However, by making thicker stems, species with low wood density can be as stable as species with high wood density. In such case, it is improper to compare construction costs and strength in stems based on wood density only, without considering the stem volume as well (e.g. King et al. 2006b).
Engineering theory has been applied to evaluate the minimum structural and biomass investment required to build a mechanically stable trunk (Niklas 1992; Anten & Schieving 2010) and horizontal branches (Sterck, van Gelder & Poorter 2006; Anten & Schieving 2010). Anten & Schieving (2010) used engineering theory to estimate measures of mechanical stability such as buckling safety factor, maximum lateral force and the resistance to rupture of branches under their own weight. By inserting data obtained at each stem and branch level from varying sources, they showed that species with low wood density require thick stems to be stable, but because of cheap volumetric construction they need less stem biomass for the same height gain compared to species with high wood density. This theoretical prediction for stems has been confirmed for intermediate-sized trees (17 m tall) of large-statured tree species in a Malaysian rainforest (King et al. 2006b) but not for saplings in an Australian subtropical forest (Kooyman & Westoby 2009). In addition, engineering theory predicts that the cost of horizontal branch expansion depends on the ratio between wet and dry wood mass, as wet mass determines self-loading and dry mass determines branch strength (Anten & Schieving 2010). Sterck, van Gelder & Poorter (2006) showed for Bolivian tree saplings that species with low wood density pay higher cost per branch length for horizontal branch extension than species with high wood density and, as a result, they produce narrower crowns than high wood density species. Anten & Schieving (2010) suggest that this relationship is likely to be restricted to saplings only, because young wood contains more water than large trees, and water does not provide strength but increases self-loading. As far as we know, there is no empirical evidence, testing the relationship between crown width and wood density throughout ontogeny across tropical forest trees (but see, Aiba & Nakashizuka 2009; for a temperate forest study). Engineering predictions for the relationships between wood density and crown depth are not straightforward. However, it can be assumed that a deep crown is advantageous because a tree thus enlarges the foliage area while the added mechanical forces on the branch and supporting stem are kept relatively low. As wood hardness increases with approximately the square of wood density (US Department of Agriculture 1999) and species with a high wood density may produce more persistent branches and gain considerable resistance to damage, they are predicted to have a relatively deep crown.
Adult stature has been found to be a major life-history trait explaining architectural differences across species based on the trade-off between vegetative growth and early reproduction (Kohyama et al. 2003; Poorter et al. 2003; Poorter, Bongers & Bongers 2006; Iida et al. 2011). It remains an open question how wood density affects the architectural differentiation throughout life history. Previous studies that evaluated the role of wood density in architectural differentiation focused only on limited ontogenetic stages, such as saplings (Sterck, van Gelder & Poorter 2006; van Gelder, Poorter & Sterck 2006) or intermediate-sized trees of large-statured species (King et al. 2006b), or on a broad range of ontogenetic stages but for few species (Sterck & Bongers 1998). Aiba & Nakashizuka (2009) showed the strong correlation between wood density and tree architecture for a broad range of sizes of 30 dominant species in a temperate forest; however, it is not clear whether the same trend can be observed in other forests and forest types. In this paper, we evaluate the relationship between wood density and tree architecture for 145 co-occurring tree species in a Malaysian rainforest and analyse size-dependent trends in stem diameter, stem biomass, height of the lowest foliage and crown width. We predict, in line with Anten & Schieving (2010), that trees with lower wood density have larger stem diameter, lower stem biomass and narrower and shallower crowns than trees of similar height with higher wood density. We also expect those relationships to be maintained when trees increase in size, if the importance of mechanical effects strongly persists across broad ontogenetic stages. The relationships observed in our study will improve our understanding of species life-history strategies and the role of interspecific trade-offs for the maintenance of species-rich tree communities.
Materials and methods
Study site and data collection
The study was conducted in a lowland mixed dipterocarp forest, the 50-ha Forest Dynamics Plot at the Pasoh Forest Reserve, Negeri Sembilan, Peninsular Malaysia (2°59′N, 102°18′E). The Pasoh plot receives a mean annual rainfall of 1810 mm (Noguchi, Nik & Tani 2003), and soils and topography of the plot are relatively uniformly flat with a maximum elevation difference of 25·5 m within the plot (Davies et al. 2003). The plot was established in 1985–1987 with standardized procedures (Condit 1998). Within the 1000 × 500 m area, all woody stems with stem diameter >1 cm at breast height (1·3 m) were tagged, mapped to the nearest 0·1 m, identified to species and measured for stem diameter to the nearest 1 mm. Subsequent censuses of growth, mortality and recruitment have been conducted approximately every 5 years. The plot contains 814 species, consisting of 295 genera and 81 families (Davies et al. 2003). We used the data of the 2000 census for selecting sampled species and 2005 census for selecting individual trees. We choose 200 tree species at random from the 423 species that had more than 100 individual trees living in the inventory for the year 2000, but for this paper we only analysed those 145 species for which wood density information was available. Twenty individual trees for each study species were selected such that they covered the large range of stem diameter from sapling to adult. We selected trees using weighted rank sampling of stem diameter (D) raised by 0·5 power of D. In this way, we avoid sampling bias towards small trees within a population that typically exhibits largely positive-skewed size distribution.
We conducted fieldwork in November–December 2006 and June–July 2007. For each of the 2900 selected individuals of the 145 species, we measured stem diameter at 1·3 m above the stem base (D), tree height (H), height of the lowest foliage (F) and two perpendicular crown widths of projected area (W1 and W2, where W1 > W2). Heights <15 m were measured using a carbon-fibre splicing pole and all others using a hand-held laser range finder (LaserAce 300 or LaserAce Hypsometer; MDL, York, England). Crown widths were determined using a steel tape, after the crown peripheral points were identified using a level. We recorded whether trees had damaged stems, crown breakage or were resprouted. In total, 2667 trees including 292 trees with damage were measured. Two hundred and thirty-three of the preselected individuals were not found in the field and were presumed to be dead. On average, 18 individuals were measured per species (range: 15–20 individuals).
Wood density data for 145 species were obtained from two sources. Wood densities for 38 species were measured by S. J. Wright and his colleagues with an increment borer just outside of the Pasoh 50-ha plot because destructive measurements are prohibited within the plot. Wood samples from 4 to 7 individuals in each species were collected and were oven dried at 60°. Trees with stem diameter more than 10 cm were selected for coring. For shrubs and treelets that have never reach 10 cm stem diameter, a standard 1-cm-diameter branch was selected (cf. Swenson & Enquist 2008) to reduce damage to the plant. The main trunk was sampled if necessary. Wood density data for 107 additional species were obtained from the global wood density database (Zanne et al. 2009; Chave et al. 2009; http://datadryad.org/repo/handle/10255/dryad.235). In this database, wood density has been measured as basic specific gravity (oven dry mass/green volume; g cm−3). Records from juveniles or from planted trees were excluded from the current database. We used wood density data obtained from South East Asia only to reduce region-specific difference as much as possible. Wood density across the selected 145 species varied from 0·32 to 0·92 g cm−3 with an average of 0·63 g cm−3 (Table S1, Supporting information). We did not account for the intraspecific variation in wood density within a tree and among trees, and consider species rankings based on average species wood densities for trees (excluding juveniles) to be valid because such rankings are consistent across databases (e.g. Chave et al. 2009). This is in agreement with the observation in Pasoh that the variation across 38 species (SD = 0·105) was larger than the variation within species (SD = 0·056) (J. S. Wright unpublished data). Wood density may vary within a stem at different positions and among individual trees with different tree sizes (e.g. Swenson & Enquist 2008; Sarmiento et al. 2011). However, the Pasoh data showed that, for adult trees, within-species variation in wood density was relatively small compared to the across-species variation. Thus, we feel confident that the observed variation in stem wood density is largely driven by the variation among (and not within) tree species (e.g. van Gelder, Poorter & Sterck 2006). To evaluate mechanical stability of the species, the safety factor for height was calculated for all intact individuals as the ratio of actual to critical buckling height (Hcr/H). Critical buckling height was calculated as Hcr = 3·83D3/2 (King et al. 2009). Species average values were then used and correlated with wood density.
Data analysis and statistical modelling of tree architecture
Architectural relationships between two variables typically rely on traditional approaches that fit a regression curve to bivariate plots of log-transformed data for a single predicted relationship. This approach ignores the fact that several architectural relationships belong potentially to a suite of interconnected associations among architectural dimensions. Another issue is that the classical methods for estimating the coefficients that describe how a particular variable of architectural dimensions scales with another variable either ignore uncertainty in one of the variables or assume a specific distribution to account for variable uncertainty (Warton et al. 2006). To address this issue, we employed a hierarchical Bayesian framework to describe the relationships among tree architectural variables, (D, H, F, W1, W2), considering variable uncertainty by applying the same procedure of Iida et al. (2011). H was assumed to be a function of D, and F and W1 were assumed to be a function of H. W2 was assumed to be proportional to W1. All parameters were interconnected in this model framework and obtained as probability distributions (hereafter, posterior distributions) at community and species level. All variables below referred to a focal tree where subscripts referring to a specific individual were omitted to simplify the notation.
As tall trees usually increase their top height, H, in asymptotic manner with D, we estimated tree height, H, using the following function:
where all parameters were positive. The parameter aH represents the asymptotic maximum tree height, and the product of aH and bH shows the initial slope of the H–D curve when D =0. The parameter dH accounts for tree damage such that an apparently undamaged trees have dH = 1 or . Damaged trees have dH < 1, and the parameter dH of damaged trees is estimated by the model.
The scaling between the two size dimensions of an organism often follows a power relationship. The scaling component indicates the ratio of the relative growth rate of one dimension to the relative growth rate of another dimension. The height of the lowest foliage, F, and the longest crown width, W1, were, therefore, approximated by power functions of H:
where aF, bF, aW and bW are allometric parameters. W2 was set as proportional to W1:
Parameters, dH and c, were estimated for the entire community, and all other parameters were estimated for both the entire community and species-specific components at log scale. Parameter descriptions are given in the Appendix S1 (Supporting information).
Sampling from the posterior distribution of all parameters was performed using the Markov chain Monte Carlo method with winbugs 1.4.3 (Spiegelhalter et al. 2003). The posterior samples were obtained from three independent Markov chains, in which a total of 1500 values were sampled with 20 iteration intervals after a burn-in of 10 000 iterations (in total 40 000 iterations). The convergence of the Markov chains was checked with (Gelman et al. 2003) for each parameter by comparing the variance within each chain and among chains.
Architectural traits and wood density
Tree architecture varies with tree size. The outcome of interspecific comparisons of architectural variables has been shown to vary substantially with the standardized size applied (hereafter, reference size), as species may show cross-over in their ontogenetic trajectories (Poorter, Bongers & Bongers 2006). To account for such size-dependent effects, we used an approach to evaluate the relationships between species architecture and wood density with increasing tree height. We calculated architectural variables (stem diameter D, stem biomass M, height of the lowest foliage F and crown width W1) at each given tree height, hereafter reference height H, using the species-specific equations. Stem biomass M was approximated as a cylindrical stem at unit height from (π/4)ρD2, where ρ is wood density and D is stem diameter at breast height for trees with reference height H estimated from species-specific H–D relationships. The correlations between probability distributions of D, M, F or W1 at reference height with wood density were calculated for increasing H, using 1-m intervals for H (cf. Poorter, Bongers & Bongers 2006; Iida et al. 2011). The same procedure was repeated as long as 20 species could be included in the comparison. To evaluate whether the observed ontogenetic patterns were because of a genuine ontogenetic trend or because of the fact that smaller species drop out of the comparison when larger reference heights are used, we selected 81 large-statured species (with 95-percentile maximum height >20 m) from the 145 species in the sample and applied the same approach explained earlier (see detail in Fig. S1, Supporting information). For all analyses, we employed the Kendall’s rank correlation. Significantly, positive or negative correlation for each pair was judged if the 95% interval of the distribution of Kendall correlation coefficient, T, did not include zero. All analyses were conducted using r 2.9.1 (R Development Core Team, 2009).
Tree architectural parameters varied considerably within species, and the 145 species showed overlapping probability distributions of tree architectural parameters. The estimated median values of posterior parameter distributions for stem diameter D, stem biomass M, crown width W and height of the lowest foliage F changed with tree height H (Fig. 1). Stem diameter increased with tree height, and some species showed an asymptote for height at large diameter classes (Fig. 1a). A similar pattern was observed for the change in stem biomass with height (Fig. 1b). The change in crown width with height was convex, while the relationship between height of the lowest foliage and height was close to linear (Fig. 1c,d).
Despite qualitatively similar interspecific trends in stem diameter, stem biomass, crown width and height of the lowest foliage with increasing tree height, the architectural variables (D, M, W and F) were correlated significantly with wood density at reference height of 5 and 25 m (except for crown width at 25 m height) and indeed, over most of the ontogenetic trajectory (Fig. 2). Wood density was correlated negatively with stem diameter (Fig. 2a–c) and positively with stem biomass (Fig. 2d–f) for most reference heights H. Thus, as we expected, species with low wood density had thicker stems, but such thicker stems were produced at lower stem biomass costs than for species with higher wood density. The safety factor for height and wood density were not significantly correlated (τ = −0·016, P-value = 0·78), indicating that there is no significant difference in mechanical stability in stems for species that vary in wood density. In crown dimensions, wood density correlated positively with crown width for H ≤18 m (Fig. 2g–i) and negatively with height of the lowest foliage for H ≤24 m (Fig. 2j–l), which suggests that species with high wood density produced wider and longer crowns than species with low wood density at small size classes living in understorey, and such trend becomes weak at large size classes when trees are close to or in the canopy.
For the subset of 81 large-statured species (with more than 20 m in adult height), we found similar relationships between wood density and architectural variables (Fig. S1, Supporting information) as found for all 145 species. This suggests that ontogenetic trends in the effects of wood density on tree architecture are genuine and not caused by smaller species dropping out of the comparison when larger reference heights are used. The observed patterns are, therefore, independent of the adult stature of the species, which is also reflected by the nonsignificant correlation between wood density and upper diameter across the 145 species (τ = −0·057, P-value = 0·31). Upper diameter is a reasonable index of adult stature and defined as 95-percentile maximum stem diameter of each species population with D >0·1 × observed maximum D in the 50-ha plot (King, Davies & Noor 2006a).
We examined the relationships between species wood density and height-dependent change in tree architecture across 145 co-occurring species in a lowland Malaysian rainforest. In line with our hypotheses based on engineering theory for mechanical stability, we found that species with lower wood density produced thicker but cheaper stems and narrower and shallower crowns than species with higher wood density. These relationships were maintained for most reference tree heights.
The role of wood density in architectural differentiation
In the context of engineering theory (Anten & Schieving 2010), our results support a mechanical role for wood density underlying architectural differences across co-occurring tree species. This is a surprising result, given that we employed a quite conservative approach that considers intraspecific variation (e.g. individual differences and measurement errors) for interspecific comparison and that we employed average wood density of species from a regional average data set. Such conservative procedures may have lead to the relatively weak correlations observed in our results.
Tree height and stem biomass
For trees of the same height, stem diameter was correlated negatively and stem biomass was correlated positively with wood density (Fig. 2a–f). This suggests that trees of lower wood density species have thicker but cheaper stems in terms of biomass than species with higher wood density, which is in accordance with the finding for Malaysian rainforest trees (King et al. 2006b), but not for Australian rainforest saplings (Kooyman & Westoby 2009). In our study, trends were maintained over most of the height range. We calculated stem biomass based on wood density and stem volume. The fact that wood density is positively related to stem biomass means that interspecific differences in wood density are much larger and a stronger driver of stem biomass than interspecific variation in stem volume. However, it is possible that light-wooded species, compared with heavy-wooded species, take higher risks, by having cheap extension costs, at the expense of a reduced mechanical stability. Yet, this is not the case, as there is no significant relationship between the safety factor for height and wood density across our 145 species in concordance with a critical result of Anten & Schieving’s (2010) engineering theory. Our study confirmed that species with low wood density can grow more efficiently in stem dimensions, both vertically and horizontally, than species with high wood density, because of the relatively low construction costs, while still being biomechanically stable.
Our results showed that high wood density species had wider crowns than low wood density species (Fig. 2g–i) in concordance with saplings in a Bolivian forest (Sterck, van Gelder & Poorter 2006) and trees at most size ranges in a temperate forest (Aiba & Nakashizuka 2009). These observations support the predictions from the engineering theory (Anten & Schieving 2010) that slender and stiff branches of high-density wood lead to efficient horizontal branch expansion per unit biomass invested. However, this relationship disappeared for trees taller than 18 m. King et al. (2006b) also found no relationship between wood density and crown width for trees of 17 m tall among Malaysian rainforest species. These relationships may disappear with larger reference heights if larger trees have wood with lower water content than smaller trees. Engineering theory predicts that in that case the effect of wood density on branch construction costs is reduced (Anten & Schieving 2010). An alternative explanation might be that in the exposed conditions of the overstorey, the mechanisms constraining tree architecture may be different from that of dark understorey. In well-lit conditions, trees start to expand their crowns to enhance their foliage area for light absorption, and under those conditions, the benefits of carbon gain may compensate for the higher branch expansion cost of low wood density species. In addition, higher wind forces on individual trees in the overstorey may cancel the potential relationship between wood density and tree architecture. Within the forest, trees show plastic responses in lateral crown expansion to their neighbourhood (Umeki 1995). Interspecific differences in tree height–crown width relationships are, therefore, relatively large (Iida et al. 2011). Such plastic responses to local light conditions may override more subtle, interspecific differences in costs of lateral crown expansion related to wood density. This might explain why our interspecific patterns with wood density are relatively weak.
Height of lowest foliage
We expected that species with high wood density would have deeper crowns than species with low wood density because branches with denser wood are more resistant to damage and disease (Loehle 1988) and may persist for a longer period. We observed that wood density was positively correlated with crown depth (Fig. 2j–l), which is in concordance with a study on 30 species in a Japanese temperate forest (Aiba & Nakashizuka 2009). A deep crown of high wood density species is consisted of dense-wooded branches, which may persist for a longer period mechanically. However, it is also necessary for keeping deep crown that branches in the bottom of the crowns have a positive carbon balance. Such a positive carbon balance may be realized for species with high wood density because they are more shade-tolerant (van Gelder, Poorter & Sterck 2006) and allow them to keep leaves with lower light compensation points or because species with dense wood may have small bark surface area of branches with low maintenance cost (Sterck, van Gelder & Poorter 2006; Larjavaara & Muller-Landau 2010). Species with wide crowns and dense wood may also display their leaves more effectively, allowing light to penetrate to lower-level branch layers. Therefore, crown depth may reflect a compromise between mechanical stability and physiological efficiency.
It is often suggested that wood density underlies the trade-off between mechanical stability and construction cost such that high wood density leads to mechanically stable stems and that low wood density leads to cheap volumetric expansion cost (e.g. Poorter et al. 2003). Yet, this trade-off does not apply to stems, because low wood density species realize the similar stability as high wood density species by making thick stems at low biomass cost as mentioned in the previous studies (Anten & Schieving 2010; Larjavaara & Muller-Landau 2010). However, when considering whole tree architecture, we found that wood density explains the trade-off between effective vertical stem expansion and horizontal crown expansion. This trade-off was manifested for trees up to 18 m tall, which grow under a forest canopy.
In tropical forests, wood density is often closely related to light demand (King et al. 2006b; Poorter et al. 2008; Aiba & Nakashizuka 2009), and therefore, the observed relationships between wood density and architecture could be also interpreted in terms of light adaptation (Falster 2006; King et al. 2006b; Aiba & Nakashizuka 2009). Across our 145 co-occurring species, wood density was indeed negatively correlated with sapling mortality in shaded conditions [τ = −0·080 with 95% CI of (−0·136, −0·023)], indicating that species with dense wood are also more shade-tolerant. However, architectural dimensions were more closely correlated with wood density (as an indicator of material properties; this study) than with sapling mortality in shaded conditions (as an indicator of light requirement for regeneration; Iida et al. 2011). This suggests that variation in wood density may result in architectural differentiation, because of biomechanical reasons.
Two axes that characterize architectural differentiation in tree communities
How is the linkage between wood density and tree architecture related to the classical theory of architecture and adult stature? For tropical tree communities in Africa (Poorter et al. 2003), South America (Bohlman & O’brien 2006; Poorter, Bongers & Bongers 2006) and Asia (Kohyama et al. 2003), including the Malaysian tree community studied here (Iida et al. 2011), it has been shown that juvenile trees of large-statured species have more slender stems and smaller crowns than similar sized trees of small-statured species. The results of the present paper indicate that species with high wood density have more slender stems, but larger crowns than similar sized trees of low wood density species. Across 145 tree species in this community, there was no significant correlation between wood density and adult stature (see also King, Davies & Noor 2006a). Similarly, no significant correlation was found between wood density and adult stature for tree communities in Kalimantan (Kohyama et al. 2003), Bolivia (Poorter et al. 2010), Australia (Falster & Westoby 2005; Kooyman & Westoby 2009), New Zealand (Russo et al. 2010), subtropical Japan (Aiba & Kohyama 1997), temperate Japan (Aiba & Nakashizuka 2009) and temperate and Mediterranean Spain (Martinez-Vilalta et al. 2010) (but see Thomas (1996) for a negative correlation for a few genera in Malaysian rainforest). Therefore, adult stature and wood density represent largely independent axes of variation that collectively characterize the interspecific differentiation in tree architecture within tree communities. Architectural differentiation based on wood density is closely related to light capture strategies: species with low wood density have an efficient height gain to attain favourable light conditions for future, whereas species with high wood density realize an efficient horizontal crown expansion for light capture at the present height, thus enhancing persistence in the shaded forest understorey (Kohyama 1987; Kohyama & Hotta 1990; Aiba & Nakashizuka 2009). Our study thus suggests that the mechanical constraints set by wood density contribute to the co-existence of species differing in architecture and light capture strategy.
We thank the Pasoh field assistants and staffs of the Forest Research Institute Malaysia (FRIM) and Tatsuyuki Seino for their kind help in field research. We acknowledge funding from ‘F. H. Levinson Fund’ and access to unpublished data from S. Joseph Wright. We gratefully acknowledge all people who are related to the Global Wood Density Database and thank Dr Niels Anten, Dr David King, and two anonymous reviewers for their comments. This study was supported by Grant-in-Aid for Scientific Research from Japan Society for the Promotion of Science (No. 19405006 & 21405006) and a grant from the Center for Tropical Forest Science. Yoshiko Iida was granted a Research Fellowship for Young Scientists and Excellent Young Researchers Overseas Visit Program from JSPS. The 50-ha Forest Dynamics Plot at Pasoh is a collaborative project of the Forest Research Institute Malaysia (FRIM), the National Institute for Environmental Studies, Japan (NIES), and the Center for Tropical Forest Science–Arnold Arboretum Asia Program, Harvard University (CTFS-AA).