Linking functional traits and demographic rates in a subtropical tree community: the importance of size dependency

Authors


Summary

  1. An important goal in plant community ecology is to understand how species traits determine demographic performance. Several functional traits have been shown to correlate with growth and mortality rates in trees, but less is known about how the relationships between functional traits and demographic rates change with tree size.
  2. We examined the associations of functional traits with growth and mortality across 43 tree species in the Fushan 25-ha subtropical rain forest plot in northern Taiwan. We estimated the 95th percentile maximum stem diameter, wood density and six leaf functional traits (leaf area, specific leaf area, thickness, succulence, and mass-based nitrogen and phosphorus contents) obtained from leaves on juvenile and adult individuals of each species.
  3. To quantify size-dependent changes in growth and mortality, relative growth rate (RGR) and mortality were estimated as a function of stem diameter using hierarchical Bayesian models. These rate estimates were then correlated with functional traits at a range of stem diameter classes.
  4. Relationships between functional traits and demographic rates varied with tree size. Maximum size was positively correlated with RGR across a wide range of tree sizes. Wood density was negatively correlated with RGR and mortality for small-sized trees. Leaf traits such as leaf area and specific leaf area at juvenile and adult stages were associated more strongly with demographic rates for corresponding sizes than from other sizes.
  5. Synthesis. The observed size-dependent changes in the trait–demography relationships are possibly due to the effects of developmental and environmental changes with increasing tree size. The underlying effects of functional traits on demographic performance vary with tree size, and this should influence dynamics in a tree community.

Introduction

The assembly and dynamics of ecological communities are often demonstrated by interspecific variation in demographic (growth and mortality) rates (e.g. Pacala et al. 1996; Rees et al. 2001). In other words, interspecific variation in demographic performance, given the environmental context, provides insights into the mechanisms underlying the co-occurrence and dynamics of species. An important component of this realm of research is to identify the morphological and physiological traits of individuals and species that are strong predictors of demographic performance. The importance of linking traits and performance has long been recognized in ecology and evolution (Arnold 1983), whereas in plant ecology, it has garnered an increased level of interest as more and more investigators have sought to quantify plant functional traits. Plant functional traits are simply morphological or physiological traits that are believed to be linked to the ecological strategies of species and their performance in a given environment (McGill et al. 2006).

An increasing number of studies have sought to explicitly link the growth and mortality rates of trees to commonly measured functional traits generally along functional axes of woody plant strategies in relation to height, stems, leaves and seeds with varying degrees of success. For example, two commonly measured functional traits related to the height and stem axes, adult stature and wood density are closely related to the demographic rates of species (Muller-Landau 2004; Poorter et al. 2008; Martinez-Vilalta et al. 2010; Wright et al. 2010; Herault et al. 2011; Rüger et al. 2012; Lasky et al. 2013; Iida et al. 2014). In particular, large-statured species tend to gain height faster at a given stem diameter than small-statured species. Therefore, trees of large-statured species tend to pre-empt light resources compared to trees of small-statured species at a similar stem diameter (Thomas 1996b; Kohyama et al. 2003; Iida et al. 2014). Further, species with low-density wood tend to grow in volume faster, but such species tend to suffer higher mortality than species with high-density wood, probably because of a reduced mechanical or physiological stability at the tissue level (e.g. xylem) and a high maintenance (respiration) cost at the organ (e.g. stem) level (Muller-Landau 2004; Chave et al. 2009; Larjavaara & Muller-Landau 2010; Iida et al. 2012). These two functional axes, adult stature and wood density, are known to independently affect demographic rates across several tree communities (Poorter et al. 2008; Martinez-Vilalta et al. 2010; Wright et al. 2010; Rüger et al. 2012; Lasky et al. 2013; Iida et al. 2014).

Given the demonstrated linkage of adult stature or wood density with tree performance, it is perhaps surprising that leaf traits are rarely significantly correlated with tree performance in these previous studies. The only study that has established these relationships in a natural forest comes from a study of saplings in a Bolivian semi-evergreen moist tropical forest (Poorter & Bongers 2006). This is surprising because we expect traits related to leaf economics (e.g. specific leaf area (SLA), leaf nutrient content), which are believed to underlie much of the variation in plant ecological strategies across ecosystems (Wright et al. 2004), to be predictive of demographic rates. We would therefore expect that species with high nutrient content and high SLA would have elevated growth and morality rates relative to species with leaves of low nutrient content and low SLA, but this prediction is generally not supported in woody plants. One possible reason why leaf traits are not good predictors of tree performance is that they must be integrated with whole-plant allocation such that a trait like SLA is integrated with the total leaf area deployed by a large plant individual (e.g. Enquist et al. 2007). A second reason, which is potentially interacting with the first reason, is that the relative importance of leaf traits for predicting performance changes with ontogeny together with size-dependent changes in leaf traits and demographic performance. The majority of published studies employ species functional traits and demographic rates for a certain size class or quantify an average trait or rate from a wide range of size classes (Poorter et al. 2008; Martinez-Vilalta et al. 2010; Wright et al. 2010 but see McMahon, Metcalf & Woodall 2011). Such averaging or aggregation could wash away leaf trait–demographic rate relationships. It is therefore still an open question whether the changes in demographic rates and/or functional traits with tree size affect their linkages. Thus, a first key challenge is to quantify the demographic rates of individuals within species at different tree sizes and to determine whether there are size-dependent shifts in the traits that best predict tree performance at a given tree size. Another key challenge for linking plant traits to performance is to consider the size-dependent changes in the functional traits themselves. This is particularly important for leaf functional traits that show a marked difference through ontogeny due to changes in the light environment and/or size dependency (e.g. Yamada & Suzuki 1996; Miyata et al. 2011). While it is logistically difficult to quantify leaf traits across all size-classes, a step forward is to quantify leaf traits in the juvenile and adult stages and to utilize this information to predict performance as it varies with tree size.

In this study, we address the above two challenges aiming to link plant functional traits with demography in a tree community. Specifically, we examine how the relationship between several functional traits and demographic rates changes with tree sizes and whether leaf traits quantified for different life stages (juvenile and adult) reveals a stronger linkage between leaf traits and demographic rates. We address these challenges by first estimating the growth and mortality rates as a function of tree size for 43 co-occurring tree species in the Fushan Forest Dynamics Plot (FDP), a 25-ha permanent plot in the subtropical rain forest of northern Taiwan. The Fushan FDP is disturbed by typhoons 0.49 times per year (Lin et al. 2011). This has resulted in a small-statured forest that is very dynamic. We examine the relationships between functional traits and demographic rates across a wide range of tree sizes by using species trait values for maximum tree size, wood density and leaf traits (specific leaf area, mass-based nitrogen and phosphorus content, thickness and succulence). Finally, these analyses were followed by an analysis of the linkage between leaf traits and size-dependent demographic rates using leaf traits quantified from either juvenile or adult individuals. The two central questions we ask in this study are (i) do the relationships between functional traits and demographic rates change with tree size? In particular, it may be expected that traits like maximum size and wood density are consistently related to demographic rates, while leaf traits may only be strongly linked to demographic rates in smaller individuals (Poorter & Bongers 2006); and (ii) do leaf traits quantified from trees at different life stages (juvenile vs. adult) reveal a stronger association with demographic rates of similarly sized trees? In particular, we expect that if leaf traits are good predictors of demographic rates along a possible trade-off between ‘grow fast – die early’ vs. ‘grow slowly – die late’, it is more likely that leaves from similarly sized trees will be more closely linked to demography than leaves from individuals in other tree sizes.

Materials and methods

Study Site and Species

This study was conducted in the Fushan FDP, a 25-ha permanent plot in the subtropical rain forest of northern Taiwan (24°45′40″N, 121°33′28″E) (Su et al. 2007). The climate of Fushan is influenced by the north-east monsoon in winter and by frequent typhoons during summer and autumn. The Fushan forest receives a mean annual rainfall of 4271 mm with an average temperature of 18.2 °C and a mean relative humidity of 95.1% (Su et al. 2007). The Fushan FDP is located in a small upstream watershed, and the topography is uneven and rugged with the elevation ranging from 600 to 733 m above sea level. The FDP was established in 2003–2004 where all woody stems within the 500 × 500 m area with stem diameter ≥ 1 cm at breast height (1.3 m) were tagged, mapped to the nearest 0.1 m, identified to species and had their stem diameter, D, measured to the nearest 1 mm. The second census of growth, mortality and recruitment was conducted during 2008–2009.

A total of 110 woody species, consisting of 67 genera and 39 families were recorded in the Fushan FDP. For this study, we used 43 tree species for which we had sufficient information regarding four leaf traits for both juvenile (1 cm ≤ D < 3 cm) and adult trees (D > D950.1) (Table S1 in the Supporting Information). The index, D950.1, represents the potential maximum size for a species, which was defined as the upper 95th percentile stem diameter, D, for trees of the subpopulation whose D was equal to or > 10% of the observed maximum D of a population (King et al. 2006b). It is important to note that this is an estimate of the maximum size in a population and that larger individuals likely occur in other populations in the geographical range of the species. We applied D950.1 based on D as the potential maximum size rather than tree height in this study (e.g. difference between juvenile and adult trees based on D and, growth and mortality at different D). This method allows us to choose a realistic species set for relationships between functional traits and demography at different D, because species with small maximum size need to be removed from the comparison with increasing D which they cannot attain. Since tree height can be scaled by stem diameter (e.g. Kohyama et al. 2003; Iida et al. 2011), we argue that this index is proper to show the potential maximum size of the species. Further, this method allows us to minimize the effects of population size and tree size structure and reduces the probability of underestimating the potential maximum size for a specific population with a large proportion of small individuals. We estimated D950.1 using the stem diameter data set from the 2008 to 2009 census. Note that D950.1 values of selected 43 species were larger than 3 cm. We used 26 of the 43 species for two leaf chemical traits due to the lack of data.

Functional Traits

Leaf traits

We collected 1–3 intact and exposed leaves or leaflets from the outer canopy of the crown for individual trees at juvenile (1 cm ≤ D < 3 cm) and adult (D > D950.1) stages for 43 species (for each stage, 1–12 trees per species) and measured leaf area (LA), specific leaf area (SLA), leaf thickness and succulence according to Cornelissen et al. (2003). The collected leaves were kept in a Ziploc bag with wet tissue and stored in a cool box in the field until they arrived at the laboratory. Thickness (mm) was measured using a dial thickness gauge (dickenmesser; Mitutoyo, Kawasaki, Japan) in the laboratory. We measured fresh leaf weight with an electronic balance to the nearest 0.1 mg. Leaf area was measured with a flatbed scanner in the laboratory within 12 h of collection. Leaves were then oven-dried for 72–96 h at 60 °C until a constant weight was reached. Leaf area (LA; cm2) was calculated using image-J (http://rsb.info.nih.gov/ij/) based on the scanned images. Specific leaf area (SLA; cm2 g−1) was calculated as leaf area per dry leaf mass. Leaf succulence (gH2O per cm−2) was calculated as (leaf wet mass – leaf dry mass)/leaf area. Then, we estimated mean values of leaves or leaflets at juvenile and adult stages for each species. For 26 of the 43 species, total organic nitrogen mass per unit leaf mass (Nmass,%) and total organic phosphorus mass per unit leaf mass (Pmass, %) were determined by two microplate methods (Huang et al. 2011, see detail in Appendix S1).

Wood density

The procedures for measuring wood density followed the Center for Tropical Forest Science (CTFS) wood density measurement protocol (http://www.ctfs.si.edu/data/documents/Wood_density_draft.pdf). Specifically, we randomly selected five individuals of each species from outside the 25-ha FDP then used increment borers to extract wood cores with a 5.1 mm inner diameter. We used increment borers only for trees larger than 6 cm in diameter at breast height. Wood cores were extracted at 1.3 m above the ground on the north facing side of the stem. The length of each core was half the diameter (to the pith of the stem). Wood cores were broken into 5 cm segments. We then used the water displacement method to measure fresh wood volume within 24 h after collection, then oven-dried each segment to a constant weight at 80 °C. The wood density for each segment was calculated using the dry weight divided by fresh wood volume. The wood density for each individual tree was calculated by the weighted mean of wood density for each segment where the weighting represented the proportion of stem cross-section area in each successive concentric ring represented by each wood core segment. Wood density across the selected 43 species varied from 0.29 to 0.79 g cm−3 with an average of 0.54 g cm−3.

To examine relationships between functional traits, we utilized Kendall's rank correlations. For relationships of leaf traits between juvenile and adult stages, we also tested whether the slope of a standardized major axis was equal to one when the intercept of the regression was zero to quantify the difference of leaf traits in juvenile and adult stages. This was done using the slope test function in the smatr R-package (Warton et al. 2012).

Size-Dependent Growth and Mortality Rate

To consider size dependency in growth and mortality rates and the uncertainty of field observation data, we estimated relative growth rate (RGR) and mortality rate by applying a hierarchical Bayesian model based on Iida et al. (2014). Because RGR usually declines with tree size (Ryan, Binkley & Fownes 1997; : Mencuccini et al. 2005; Rose et al. 2009), we applied a linear function to estimate RGR. In the model, it is assumed that the RGR of ith individual tree, Ri, was a linear function of the natural logarithm of the initial stem diameter D1i of individual tree i with parameters for species j (rkj, k = 1,2) which were obtained as probability distributions at the community and species level.

display math(eqn 1)

The logarithm of the final stem diameter D2i was assumed to be the sum of the logarithm of initial stem diameter D1i and the product of Ri and the census interval of tree i, t2,it1,i, as ln(D2i) = ln(D1i) + Ri(t2it1i). We calculated growth rates for 63 494 individual trees belonging to 43 species, which were alive in 2003 and 2008, and estimated the probability distribution of the parameters r1j and r2j.

To describe nonlinear size-dependent changes in mortality (e.g. U-shaped curve with size; e.g. King, Davies & Noor 2006a), we developed a hierarchical Bayesian model of mortality based on Rüger et al. (2011). Our mortality model is based on the observation of a tree individual i and whether it survived through the census period (Si = 1) or not (Si = 0). We assumed that Si followed a Bernoulli distribution with the predicted probability of survival pi as Si ~ Bernoulli (pi). Survival probability (pi) of the ith individual tree was calculated from an instantaneous mortality rate (Mi, year−1) of an individual tree i and adjusted for the time period between the first census (t1i) and the second census (t2i) as pi = exp[−Mi(t2i–t1i)]. Mi was predicted as a function of the stem diameter of individual i at the first census, D1i:

display math(eqn 2)

Parameter m1j shows the initial mortality rate, and m2j and m3j show the effect of D or ln(D) on mortality. We used census data of 73 253 individual trees that were alive in 2003 and recensused in 2008 to estimate the probability distribution for each of the parameters.

Full descriptions of the models are given in Appendix S2. Sampling from the probability distribution of all parameters in models of RGR and mortality was performed using the Markov chain Monte Carlo method in winbugs 1.4.3 (Spiegelhalter et al. 2003).

Relationships between Functional Traits and Demographic Rates

To demonstrate the size-dependent changes in relationships between demographic rates and functional traits, we calculated Kendall's correlation coefficient, τ, for all combinations between estimated demographic rates (R and M), D950.1, WD and leaf traits (LA, SLA, thickness, succulence, Nmass and Pmass) at the juvenile and adult stages. We estimated the probability distributions of RGR (R) and mortality rate (M) at a given stem diameter (reference D) in 1 cm intervals from 1 to 22 cm which included 20 species based on D950.1 using eqns 1 and 2 (cf. Iida et al. 2014). For leaf traits at juvenile and adult stages, species-specific mean values for each stage were used. Because parameters of our R and M models were estimated as probability distributions at each reference D (Table S2), correlation coefficients were also determined as a probability distribution at each reference D. We considered Kendall's correlation to be significant if the 95% interval (from 0.025 to 0.975 quantiles) of the probability distribution of the coefficient τ did not include zero. To eliminate the confounding effect of D950.1 or WD on leaf traits and demographic rates, we applied Kendall's partial rank correlation based on relationships between leaf traits and D950.1 or WD shown in Table 1 (Siegel and Castellan 1988). This allows us to examine the independent effects of D950.1 or WD on growth or mortality because D950.1 and WD were closely related with some of the leaf traits (Table 1) and R and M (Fig. 3). The number of species used in the correlation analysis decreased with reference D from 43 to 20 for morphological traits, due to the dropping out of small-statured species. Species were included in the correlation analysis when their D950.1 was larger than each reference D. To clarify the effect of species number, which decreased with increasing reference D, we selected the 20 largest statured species based on their D950.1 and applied the same correlation analyses as that used for all 43 species. All statistical analyses were conducted using the statistical computing program R 2.14.2 (R Development Core Team 2012).

Table 1.  Kendall's correlation coefficient between leaf traits, 95th percentile maximum stem diameter (D950.1) and wood density (WD)
Leaf traitsLife stageD950.1WD
τP-valueτP-value
  1. a

    < 0.05.

  2. b

    P < 0.01.

LA (cm2)Adult−0.150.162−0.26a0.015
Juvenile−0.200.062−0.23a0.032
SLA (cm2 g−1)Adult−0.27a0.010−0.26a0.012
Juvenile−0.140.203−0.33b0.001
Thickness (mm)Adult0.070.5460.120.260
Juvenile0.000.9830.090.405
Succulence (gH2O cm−2)Adult−0.160.138−0.130.243
Juvenile−0.22a0.040−0.170.117
Nmass (%)Adult−0.30a0.033−0.180.203
Juvenile−0.28a0.047−0.210.135
Pmass (%)Adult−0.110.415−0.180.193
Juvenile−0.240.085−0.33a0.017

Results

Relationships between Functional Traits

In Table 1, we illustrated the relationships between 95th percentile maximum size (D950.1), wood density (WD) and leaf traits. The correlation between D950.1 and WD was not significant (τ = 0.08, P = 0.45). Leaf area (LA) and specific leaf area (SLA) were negatively correlated with wood density in both adult and juvenile trees. For leaf chemical traits, only Nmass was negatively correlated with D950.1 in both adult and juvenile trees. The correlation between juvenile and adult leaf traits was strongly significantly positive for all six leaf traits (Fig. 1). For SLA, the slope for the relationship between juvenile and adult trees was significantly lower than one, suggesting that SLA in adult trees tended to be smaller than that in juvenile trees (Fig. 1b). In contrast, the slopes for leaf thickness and succulence are significantly higher than one (Fig. 1c,d). These indicate that leaves are thinner with less water at the juvenile stage than at the adult stage. The LA, Nmass and Pmass values did not significantly differ between juvenile and adult stages and were distributed along the 1:1 lines (Fig. 1a,e,f).

Figure 1.

Relationships between leaf traits at juvenile and adult stages, (a) leaf area, LA, (b) specific leaf area, SLA, (c) leaf thickness, (d) leaf succulence, (e) mass-based nitrogen content, Nmass and (f) mass-based phosphorus content, Pmass. Correlation coefficient, tau, was shown on upper left at each panel, and asterisk show difference in P-value by Kendall's rank correlation test (**P < 0.01, ***P < 0.001). Each line shows 1:1 relationship. We applied slope test if the slope of a standardized major axis equals to 1 when a intercept is zero and showed estimated slope and their 95% confidence interval (CI) for the slope as estimated slope [95% CI] with P-value (NS; non-significant, **P < 0.01, ***P < 0.001).

Relationships between RGR and Functional Traits

Relative growth rate (RGR) declined with increasing tree size for most species (Fig. 2a). Further, D950.1 was positively correlated with RGR over the whole range of D from 1 to 22 cm (Fig. 3a), indicating that large-statured species tend to have higher RGR across a wide range of tree sizes. Wood density (WD) was negatively correlated with RGR only from 6 to 12 cm stem diameters (Fig. 3c), indicating that species with low wood density tend to have high RGR at intermediate sizes.

Figure 2.

Size-dependent changes in RGR and mortality rate across 43 species. Lines show changes in RGR (a) and mortality (b) with increasing stem diameter, D, by applying medians of probability distributions of RGR and mortality for 43 species.

Figure 3.

Size-dependent changes in correlation coefficients between maximum size, wood density and RGR/mortality. Partial rank correlation was applied to eliminate the effect of 95th percentile maximum size or wood density on growth and mortality. Circles indicate median values of correlation coefficients. In the case that 95% interval of probability distribution of tau does not include zero, the correlation between traits and RGR/mortality is significant, and then circle was filled. The downward-sloping line shows the decline of number of species compared.

Leaf area (LA) and specific leaf area (SLA) were both positively correlated with RGR in adult trees and negatively correlated with RGR in juvenile trees (Fig. 4a,b). This indicated that species with small LA and SLA tend to have higher RGR at small sizes, and species with large LA and SLA tend to have higher RGR at relatively large sizes. In contrast, leaf thickness and succulence at both the juvenile and adult stages were negatively correlated with RGR across trees for most sizes (Fig. 4c,d). These results indicated that species with thin leaves and/or less leaf water content tended to grow faster than species with thick leaves and/or higher leaf water content. Both leaf chemical traits, Nmass and Pmass, were positively correlated with RGR at the adult stage, and the correlation became significant for large-sized trees (Fig. 4e,f). Interestingly, Pmass at the juvenile stage was negatively correlated with RGR but was positively correlated with RGR throughout trees for most sizes at the adult stage (Fig. 4f). For the 20 largest statured species (or the 15 largest statured species for leaf chemical traits) for which the species number did not change with increasing reference size, correlations between leaf traits and RGR changed with increasing size (Fig. S1). This shows that size-dependent change in correlations was not only due to decrease in species number but also due to the size-dependent change in RGR as well as in leaf traits.

Figure 4.

Size-dependent changes in correlation coefficients, tau, between leaf traits and RGR. Correlations between species leaf traits and RGR were examined for leaf traits at different life stages (juvenile; triangle or adult; circle) and for RGR at different reference stem diameter, D, at 1-cm intervals. Each value of correlation coefficient indicates the median of probability distribution of correlation coefficient, tau. In the case that 95% interval of probability distribution of tau does not include zero, the correlation between leaf traits and RGR is significant, and then circle or triangle was filled. Forty-three species were applied for 4 leaf morphological traits (a) leaf area, LA, (b) specific leaf area, SLA, (c) thickness and (d) succulence. 26 species were applied for two chemical traits (e) mass-based nitrogen content, Nmass and (f) mass-based phosphorus content, Pmass. The downward-sloping line shows the decline of number of species compared. Species whose D950.1 is less than reference D was excluded from comparison. Partial correlation was applied to exclude effect of D950.1 or wood density based on correlations shown in Table 1 (see detail in method).

Relationships between Mortality and Functional Traits

Species mortality varied across species with increasing tree size (Fig. 2b). Wood density was continuously negatively correlated with mortality at sizes < 12 cm stem diameter, but D950.1 was significantly negatively correlated at only large tree sizes (Fig. 3b,d). This indicates that species with dense wood tend to have low mortality at small tree sizes and relatively large-statured species tend to have low mortality for large-sized trees.

Leaf area (LA) at both the juvenile and adult stages was positively correlated with mortality at small tree sizes < 5 cm stem diameter, but the correlations became non-significant for LA at the juvenile stage and became negative for LA at the adult stage when tree size was more than 15 cm in stem diameter (Fig. 5a). This indicates that species with large leaf area at both juvenile and adult stages have relatively high mortality at small tree sizes, but species with large leaf areas at adult stages exhibit low mortality at large tree sizes. Specific leaf area (SLA) at both juvenile and adult stages was significantly positively correlated with mortality only for small trees < 8 cm stem diameter, but correlations of SLA in juvenile trees were stronger than those in adult trees (Fig. 5b). This indicates that species-specific SLA was only related to mortality at small tree sizes and species with large SLA tended to have high mortality at small sizes. Leaf thickness at juvenile and adult stages was negatively correlated with mortality only at small tree sizes < 7 cm stem diameter (Fig. 5c). This indicates that species with thick leaves at both adult and juvenile stages tend to have low mortality rates at small sizes. Succulence at both juvenile and adult stages was not correlated with mortality for most tree sizes (Fig. 5d). Both Nmass and Pmass at both juvenile and adult stages were significantly positively correlated with mortality at small size trees (Fig. 5e,f). A few significant correlations were found for the 20 largest statured species between leaf traits and mortality, but a shift from significant to non-significant correlations was found with increasing size (Fig. S2). This shows that size-dependent changes in correlations were not only due to the decrease in species number but also due to the size-dependent change in mortality and leaf traits.

Figure 5.

Size-dependent changes in correlation coefficients, tau, between leaf traits and mortality. Correlations between species leaf traits and demographic rates were examined for leaf traits at different life stages (juvenile; triangle or adult; circle) and for mortality at different reference stem diameter, D, at 1-cm intervals. Each value of correlation coefficient indicates the median of probability distribution of tau. In the case that 95% interval of probability distribution of tau does not include zero, the correlation between leaf traits and mortality is significant, and then circle or triangle was filled. Forty-three species were applied for 4 leaf morphological traits (a) leaf area, LA, (b) specific leaf area, SLA, (c) thickness and (d) succulence. Twenty-six species were applied for 2 chemical traits (e) mass-based nitrogen content, Nmass and (f) mass-based phosphorus content, Pmass. The downward-sloping line shows the decline of number of species compared. Species whose D950.1 is less than reference D was excluded from comparison. Partial correlation was applied to exclude effect of D950.1 or WD based on correlations shown in Table 1 (see detail in method).

The correlation coefficient between RGR and mortality was always positive in all size classes, but only significantly correlated for trees from 4 to 6 cm stem diameter (Fig. 6). This suggests that a trade-off between high growth and high mortality vs. low growth and low mortality occurred in most species in Fushan forest but that this trade-off was only significant when trees were very small.

Figure 6.

The size-dependent change in correlation coefficients between RGR and mortality. A circle indicates the median of probability distribution of tau between RGR and mortality. In the case that 95% interval of probability distribution of tau does not include zero, the correlation is significant, and then circle was filled. The line indicates the decreasing of number of species compared at each reference D from 43 to 20 species.

Discussion

The assembly and dynamics of species in communities are the consequence of intra- and interspecific variation in individual performance (e.g. growth and mortality) in response to the environment. A key challenge in ecology has therefore been to identify those aspects of form and function in several organs of the organism that underlie the performance at the level of individuals and species (Arnold 1983). Here, we argue that an important aspect of this study that has been generally left uninvestigated is the significance of size dependency in demography and functional traits themselves.

To address the issue of size dependency, we first estimated growth and mortality rates across a wide range of tree sizes for 43 tree species in a subtropical rain forest in Taiwan (Fig. 2) and related these demographic rates to species-level functional trait values for maximum size, wood density and leaf traits. We showed that interspecific variation in relative growth rate was positively correlated with mortality rate in accordance with previous studies (Poorter et al. 2008; Wright et al. 2010; Iida et al. 2014). We also provided clear evidence that this demographic trade-off between growth and mortality is consistent across a wide range of tree sizes, but it was significant only when trees were small (Fig. 6).

The question still remains about what causes such size-dependent change in demographic rates. Relative growth rate (RGR) declines with increasing tree size (Fig. 2a) and has been attributed in other studies to ageing, increased respiration costs, a lower ratio of leaf area per unit living biomass, increased self-shading and increased reproductive allocation (e.g. Mencuccini et al. 2005; Rose et al. 2009; Iida et al. 2014). Mortality rate also decreased with increasing tree size, but for some species, mortality increased at larger sizes and showed a U-shaped curve (Fig. 2b). This is probably due to a change in local environmental conditions (e.g. light), increased water limitation with height and a shift in resource allocation between vegetative growth and reproduction (e.g. Thomas 1996a; King, Davies & Noor 2006a; King et al. 2006b; Iida et al. 2014). We expect that some functional traits in part underlie such size-dependent change in demography, and therefore, we quantified their correlations and size-dependent patterns.

Our results indicated that the relationships between functional traits and demographic rates varied with tree size (Figs 3-5). For example, wood density was negatively related to RGR and mortality only when trees were small (Fig. 3c,d). Thus, our findings showed that the association between high wood density, and low growth and low mortality suggested by previous studies (Poorter et al. 2008; Martinez-Vilalta et al. 2010; Wright et al. 2010; Rüger et al. 2012; Lasky et al. 2013; Iida et al. 2014) is not always supported when the effect of tree size differences is considered in the analysis. This result could potentially be due to indiscriminate mortality due to typhoon disturbance. Previous work has shown that hurricane-induced mortality may be weakly, but not significantly, negatively related to wood density (Zimmerman et al. 1994). However, our finding that wood density was correlated with mortality only for small-sized trees under the canopy is similar to that found in a recent study using a similar analytical approach in a lowland rain forest in Peninsular Malaysia which does not experience frequent large-scale disturbance such as typhoons/hurricanes (Iida et al. 2014). Thus, it may be that when light is not limiting, the wood density–demographic relationship is less prominent.

Maximum size was positively correlated with RGR across all tree sizes, but it was generally a poor predictor of mortality rates except for some large-sized trees (Fig. 3a,b) which is in concordance with previous studies (Poorter et al. 2008; Martinez-Vilalta et al. 2010; Wright et al. 2010; Iida et al. 2014). Thus, the trend that large-statured species grow faster at a given tree size (in this study, stem diameter) than small-statured species has been confirmed across a wide range of size classes and among several forest types (Kohyama et al. 2003; Poorter et al. 2008; Wright et al. 2010; Martinez-Vilalta et al. 2010; Herault et al. 2011; Rüger et al. 2012; Iida et al. 2014).

Previous studies have highlighted linkages between demography, wood density and maximum size (Poorter et al. 2008; Wright et al. 2010; Martinez-Vilalta et al. 2010; Herault et al. 2011; Rüger et al. 2012; Iida et al. 2014); however, most of this work has failed to uncover relationships between demography and leaf traits (but see Poorter & Bongers 2006). In this study, we focused on the possibility that the effect of leaf traits on demographic rates changes during ontogeny due to size-dependent changes in demographic rates and/or the leaf traits themselves. Therefore, we examined these predictions by using leaf functional trait values from juvenile and adult trees and estimated the statistical relationship between demographic rates (RGR and mortality) with juvenile or adult leaf traits across a wide range of tree sizes (Figs 4 and 5).

We found that each of the six leaf traits was correlated with RGR, but that the detected relationships varied with tree size and were dependent on whether a leaf trait value was from juvenile or adult trees. Consistent with our first prediction, the relationship between RGR and leaf area (LA), specific leaf area (SLA), mass-based nitrogen content (Nmass) and phosphorus content (Pmass) varied with tree size (Fig. 4). LA and SLA were negatively correlated with RGR at small tree sizes, whereas they were positively correlated with RGR at larger sizes. A possible explanation for the switch in the directionality of this relationship is the change in light and water conditions along a vertical profile in a forest (e.g. Yoda 1974). In small sizes, the negative association of LA and SLA with RGR can be understood in terms of the differentiation between shade-tolerant species in the dark forest floor and light-demanding species that are situated in light gaps and the differentiation in habitat preference between valley sites and ridge sites in relation to variation in water availability. Shade-tolerant species persist under shade by large SLA and/or LA, while light-demanding species enjoy high radiation in gaps with adaptive leaf habits of small SLA and/or LA exhibiting high growth rate. By contrast, large-sized adult trees are located in light-exposed canopy where interspecific differences in SLA and/or LA are expected to reflect differentiation in topographic habitat preference. Species with high SLA [associated with high stomatal conductance (e.g. Reich et al. 1991)] and/or large LA [enhanced by high water availability (e.g. Givnish 1979)] adapted to moist habitats in lower hill slopes that support a higher growth rate, whereas those with low SLA (with low stomatal conductance) and/or small LA (restricted by low water availability) are distributed in dry habitats of upper hill slopes that in turn depresses tree growth rate. This is supported by field observations where canopy height in the upper hill slopes is much shorter than that in the moist lower hill slopes. Such a shift in the statistical relationships between RGR and leaf traits would result in an apparent non-significant relationship if, as is often employed in the previous studies, RGR would be estimated in size-independent manner. Contrary to the results of four leaf traits of LA, SLA, Nmass and Pmass, leaf thickness and succulence were consistently negatively related to RGR, which suggests that the species-level trait values may be applicable regardless of tree sizes when examining relationships with growth rates.

The relationships between leaf traits and mortality were generally much stronger than those between leaf traits and RGR. As we predicted, the relationships between leaf traits and mortality were generally only significant in small-sized trees (Fig. 5). For example, SLA, Nmass and Pmass were strongly positively correlated with mortality rates in small sizes, but they were uncorrelated with mortality in large sizes. Thus, the possible linkage of high leaf-level performance in light capture (high SLA) and assimilation efficiency (high Nmass and Pmass) with tree-level cost of high mortality are only upheld in small-sized trees. It is likely that shade-tolerant species with large SLA still suffer high mortality due to the light-resource shortage in the dark understorey. Leaf traits are possibly linked with demographic rates of large-sized trees, but this linkage may only be demonstrated when integrating the trait values across the entire crown deployed by the individual (Enquist et al. 2007).

A second major prediction we made was that leaf functional trait values from juveniles would likely be better predictors of juvenile demography than trait values from adults, and vice versa. This prediction was based on the known ontogenetic change in light conditions an individual tree experiences and therefore a resulting change in leaf trait values with ontogeny. Our prediction was also motivated by the fact that juvenile and adult leaf traits are often only loosely correlated with each other (Fig. 1). We found that differences in juvenile and adult traits did not greatly affect the detected relationships between traits and mortality across a wide range of tree sizes (Fig. 5). In contrast, we found that the LA and SLA of juveniles were more strongly correlated with juvenile RGR than those for adults, and vice versa (Fig. 4). This was also somewhat true for Nmass and Pmass. Therefore, while in some cases, we did find that size-specific trait values improved the statistical relationship between leaf traits and RGR, in other cases there was no marked improvement. This could partly be the result of our sampling design (leaf samples from either juvenile or adult stage), but our simple sampling strategy will help future broadscale comparative research on size dependency in trait–demography relations.

Conclusions

In this work, we sought to identify whether the relationship between several commonly measured traits and demographic rates in trees depends on the size of trees. We found that trait–demography relationships covaried with increasing tree size largely due to large size-dependent variation in demographic rates. The observed size-dependent changes in the trait–demography relationships are possibly due to effects on plant function by developmental and environmental changes such as light availability along the vertical gradient in a forest structure. The underlying effects of functional traits on demographic performance vary with tree size, which should influence dynamics in a tree community.

Acknowledgements

We acknowledge Takuya Kubo and two anonymous referees for their helpful comments. Fushan FDP is supported by the Taiwan Forestry Bureau and the Taiwan Forestry Research Institute. We thank the staff at Fushan Research Center for providing logistic support, and we thank many student volunteers for their assistance with fieldwork. This study was supported by grants from National Science Council of Taiwan to I.F.S. (NSC 99-2621-B-259-002-MY3) and to J.M.C. (NSC 98-2313-B-029-001-MY3). Y. I. was granted the Japan Society for the Promotion of Science (JSPS) Institutional Program for Young Researcher Overseas Visits and JSPS Postdoctoral Fellow for Research abroad.

Ancillary