INTRODUCTION
In primary and secondary rain forests in South-East Asia and South America, bamboo species often account for a substantial fraction of the plant diversity (Bystriakova et al. Reference BYSTRIAKOVA, KAPOS, LYSENKO and STAPLETON2003, Reference BYSTRIAKOVA, KAPOS and LYSENKO2004). In tropical lowlands, bamboos are usually restricted to disturbed sites (Gagnon Reference GAGNON2009, Smith & Nelson Reference SMITH and NELSON2011, Zhang & Cao Reference ZHANG and CAO1995) or riparian habitats, and are rarely found in lowland primary rain forests. In contrast, in hill and montane forests, indigenous bamboo species are common components (Bystriakova et al. Reference BYSTRIAKOVA, KAPOS, LYSENKO and STAPLETON2003, Reference BYSTRIAKOVA, KAPOS and LYSENKO2004; Dransfield Reference DRANSFIELD1992, Wong Reference WONG1995). Bamboos that coexist with woody trees in these forests occur over a wide range of light conditions, from dark understorey to fully lit clear-cut sites.
Since bamboos seldom reach the tall top canopy layer of tropical rain forests (Widmer Reference WIDMER1998, Wong Reference WONG1995), their persistence depends on the ability to survive in the low-light understorey or to colonize high-light canopy gaps. However, unlike gap-dependent tree species, bamboos often lack effective seed dispersal mechanisms as well as dormancy (Bellairs et al. Reference BELLAIRS, FRANKLIN and HOGARTH2008, Janzen Reference JANZEN1976, Wong Reference WONG1995), which would limit their ability to exploit gaps by sexual reproduction. Many bamboo species thus dominate tropical rain forests after disturbances by means of vegetative propagation. They increase clump (genet) size by recruiting culms (ramets) via rhizome branching. The clumped genet structure of tropical bamboos, however, limits their ability to exploit gaps by horizontal vegetative growth. There remains an open question about what exact mechanisms enable bamboos to establish in canopy gaps and to survive under a closed canopy (Swaine & Whitmore Reference SWAINE and WHITMORE1988). Previous studies hypothesized that the capability of bamboos to exploit a wide range of light conditions is explained by plasticity in leaf traits (Montti et al. Reference MONTTI, VILLAGRA, CAMPANELLO, GATTI and GOLDSTEIN2014, Yang et al. Reference YANG, SUN, ZHANG, COCHARD and CAO2014) and in genet architecture (Widmer Reference WIDMER1998) since photosynthetic capacity of bamboos does not differ substantially from co-occurring woody trees (Bai et al. Reference BAI, JIANG, CAO, LIAO and WAN2011, Dierick et al. Reference DIERICK, HÖLSCHER and SCHWENDENMANN2010, Saha et al. Reference SAHA, HOLBROOK, MONTTI, GOLDSTEIN and CARDINOT2009).
An alternative hypothesis is that bamboos’ hollow culm and modular growth contribute to efficient light capture across a wide range of light conditions. Bamboo culms are hollow with internal cross walls that require less biomass to attain a given height and are robust against bending as compared with solid stems. Such hollow stem architecture is impossible for woody trees as they extend stem-wood area by peripheral cambial growth. In addition, construction of bamboo culms relies mostly on subsidized carbohydrates from other ramets within a genet for rapid elongation of culms (Kleinhenz & Midmore Reference KLEINHENZ and MIDMORE2001). Bamboos, thus, maintain their crowns dynamically by replacing short-lived ramets that do not grow after initial shoot expansion (Franklin et al. Reference FRANKLIN, PRIOR, HOGARTH and MCMAHON2010).
In this study, we tested the specific hypothesis that the hollow culms and modular architecture of bamboos are more efficient in light capture than woody trees so that photosynthetic gain of an individual ramet compensates for the construction costs of short-lived ramets allowing for the quick regulation of their clump structures as light availability changes. We tested this hypothesis by quantifying the architecture and biomass dynamics of two dominant clumping bamboo species, Gigantochloa ligulata and Schizostachyum grande, in comparison with those of co-occurring trees, in a hill dipterocarp forest in Peninsular Malaysia. Given that the light intensity declines from canopy surface to understorey, the crown foliage height determines available light resources. We evaluated the efficiency of light capture by quantifying above-ground architecture of bamboos and woody trees as a function of total above-ground biomass and crown height. Finally, we estimated biomass using allometric equations, and estimated demographic biomass turnover rates based on repeated censuses of permanent plots set along a disturbance gradient created by experimental timber harvesting.
METHODS
Study site
We carried out field research at Temenggor Forest Reserve in Perak State, Malaysia. The forest reserve is located in the northern part of Titiwangsa Mountain Range in Peninsular Malaysia. The study site covered 200 ha of the reserve (600–800 m asl with average slope of 27.5°), with the main ridge running from east to west dividing the southern and northern slopes (Appendix 1). Annual rainfall at the study site recorded by the Forest Research Institute Malaysia (FRIM) in 2011 was 2215 mm, which was in line with the long-term range of 2000–3000 mm y−1 observed in the town of Gerik, 80 km west from the study site (Azmi & Sukumaran Reference AZMI and SUKUMARAN2002). Rainfall exhibited moderate seasonality; November to March experienced more rain than the other months (Azmi & Sukumaran Reference AZMI and SUKUMARAN2002). Primary upper hill dipterocarp rain forest, which typically has 50-m-tall emergent trees and 30–35-m-tall continuous canopy surface, characterized the forest reserve.
Permanent study plots were established in 2008 as a part of an experimental logging project led by FRIM. The reduced-impact logging was conducted by Perak Integrated Timber Complex, a state-owned logging company (Yap Reference YAP2010), which created a wide gradient of canopy conditions from intact to almost clear cut. Prior to timber harvesting, FRIM established 24 rectangular plots each 20 m wide by 80 m long (3.84 ha in total) distributed on both sides of slopes in the study site (Appendix 1). A census of all the individual trees with D tree ≥ 5 cm (Table 1) in 2008 recorded 3212 stems representing 422 woody tree species (65 families) in 3.84 ha and a basal area of 36 m2 ha−1. Timber harvesting was carried out in 2010. As a result of a planned timber harvesting experiment, most of the 24 plots experienced partial loss of canopy cover. Approximately 12 mo after the tree harvesting, in 2011 and 2012, we conducted field observations of bamboos.
Table 1. Notation list of architectural dimensions, biomass turnover attributes and stand scale properties.
Studied bamboo species
Four bamboo species occurred within the permanent plots. We chose the two most abundant species in terms of ramet density (with culm diameter at breast height, D ramet ≥ 2 cm), Gigantochloa ligulata Gamble (1107 ramets ha−1), and Schizostachyum grande Ridl. (556 ha−1), to be the target species for this study. These species are endemic to the Malay Peninsula, and are commonly distributed in hill and upper hill forests (Wong Reference WONG1995). Gigantochloa ligulata was the shortest among the four co-occurring bamboo species with 6–9-m-tall genet crowns, and Schizostachyum grande was the tallest bamboo in the study site with 9–15-m-tall crowns (Wong Reference WONG1995). According to the monograph of Malaysian bamboos by Wong (Reference WONG1995), the two species commonly occur in disturbed areas with G. ligulata found from lowlands to hillsides, and S. grande more restricted to hillsides. Schizostachyum grande exhibits a monocarpic flowering pattern, a common property of bamboos, meanwhile G. ligulata exhibits continuous partial flowering among ramets in an individual genet. Neither of the species showed synchronized flowering among genets during the course of our study.
Measurement of crown architecture
We sampled ramets of these two bamboo species outside the 24 plots in the study site (Appendix 1). Both species exhibited sympodial rhizome branching and formed distinct clumps consisting of multiple ramets that were closely packed at the ground level. We assumed that each visible clump of culms was an independent genet, though exceptions have been shown to exist for other clumping species (Franklin et al. Reference FRANKLIN, KANEKO, YAMASAKI and ISAGI2008). We selected 29 and 25 genets of G. ligulata and S. grande, respectively, with six to eight genets each in the following four size classes. In order to sample genets from a wide range of genet sizes, the genet size class was determined based on the observed frequency distributions of ramet number per genet (n) on a logarithmic scale in the permanent plots, where only ramets with culm diameter at breast height ≥ 2 cm (D ramet) were counted. The size classes were 2–5, 6–15, 16–36 and 37–83 ramets per genet for G. ligulata, and 2–4, 5–9, 10–21 and 22–45 ramets for S. grande.
For each of the selected genets, we calculated the total basal area of all constituent ramets (S genet, cm2) and measured the top crown height (H genet, m) with either a splice measurement pole up to 15 m (AT-15, Myzox) or an electronic hypsometer (VERTEX IV, Haglöf Sweden AB, Långsele) (Appendix 2). We measured two perpendicular horizontal crown widths along the contour and slope directions, from which we obtained the area of leafy crown (C genet, m2) assuming ellipsoids (Appendix 2). From each selected genet, we measured the top height of a ramet (H ramet, m) that had an intermediate diameter of the ramets in that genet, and then harvested the ramet from the base at ground level.
Since each internode of a culm is a modular unit bearing branches and leaves (McClure Reference MCCLURE1966, Pearson et al. Reference PEARSON, PEARSON and GOMEZ1994), we used internodes within a culm, including the lower node, as a sampling unit from which we estimated the architectural dimensions of a whole ramet. We sampled every other internode for G. ligulata and every fourth internode for S. grande (because G. ligulata showed more irregular branching on a culm than S. grande). We measured fresh weight of culm, branch and leaves in the field, and brought back partial subsamples of culm-stem and leaves (two from every sampled internode) from base to top of internode samples of every culm to the laboratory in FRIM. There they were oven dried at 80°C over 4 d and then weighed. We estimated the oven-dry weight of components of sampled internodes by applying the ratio of dry to fresh weight of subsamples (culm-stem ratio was used for culm stem and branch stem). We measured fresh leaf area of leaf subsamples using a photo scanner (LiDE 110, CANON, at 75 ppi) and image analysis software (Image J, NIH). We used the ratio of leaf area to fresh leaf weight of each sampled internode to estimate the total leaf area per internode. Finally, we estimated biomasses of culm (W culm, kg), branch (W branch, kg), and leaf (W leaf, kg), and total leaf area (A ramet, m2) in each culm using sampled internodes with the equation:
$$\begin{equation}
{P_{{\rm{ramet}}}} = {\rm{ }}[{k_{{\rm{total}}}}/{k_{{\rm{sample}}}}]{\rm{ }}{\Sigma _i}{P_i}
\end{equation}$$
where P ramet is a ramet-scale dimension (biomass or area), k total is the number of internodes in the culm, k sample is the number of internodes sampled, and Pi is an internode-scale dimension of the i-th internode sample.
In the permanent plots, we measured the crown architecture of bamboos at the first census. To determine genet-scale top height (H genet), total basal area of all culms (S genet) and crown area (C genet) (Appendix 2), we surveyed all genets in the central 10 × 30 m of the 24 plots (Appendix 3). In total, we obtained data from 97 genets for both G. ligulata and S. grande.
Bamboo allometries
We employed a set of allometric relationships to link architectural dimensions at the ramet-scale (D ramet, W culm, W branch, W leaf and A ramet) and those at the genet-scale (n, S genet, C genet and H genet) (Figure 1; Appendix 2). To estimate allometric parameters of a Y-on-X allometry for bamboos, we employed power functions, Y = exp(β)Xα , where Χ and Y are architecture dimensions, α is a scaling exponent, and β is an allometric constant. As suggested by Wright et al. (Reference WRIGHT, FALSTER and WESTOBY2006), ordinary linear regression analysis is appropriate, when the purpose is to predict a dimension from another explanatory dimension. Therefore, we estimated α and β by fitting a log-linear fixed model, ln Y = β + α ln X, with Type I regression, i.e. measurement errors are assigned on Y measures. We examined species differences in α or β by setting species identity as a fixed effect, and conducted model selection based on the corrected Akaike information criterion, AICc (Hurvich & Tsai Reference HURVICH and TSAI1989). This allowed us to examine whether (1) both α and β were different between species, (2) only β was species specific, or (3) both α and β were common across species. We estimated genet-scale dimensions of biomass (W genet) and leaf area (A genet) by applying the ramet-scale allometries to the observed D ramet of all the constituent ramets of each genet (Figure 1).
Figure 1. Allometric estimation for bamboo ramets, bamboo genets and woody trees in a Malaysian hill rain forest. The type of arrow indicates allometric equations that connect dimensions. Enclosed allometric equations for trees are by Kato et al. (Reference KATO, TADAKI and OGAWA1978) in Pasoh Nature Reserve. See Table 1 for the definition of the dimensions and Appendix 2 for observation scheme.
Tree architecture and allometry
For woody trees, we measured stem diameter at breast height (D tree, cm), tree height (H tree, m) and the area of leafy crown (C tree, m2) for all individuals with D tree ≥ 5 cm in eight plots (1.28 ha) of the 24 permanent plots (20 × 80 m each; Appendices 1, 3) in 2008, using the same protocol for bamboos. We employed a hyperbolic function for H tree on D tree allometry (1/H tree = 1/(βD tree α) + 1/γ) and a power function for C tree on D tree allometry (C tree = exp(β)D tree α) (Figure 1).
To calculate above-ground tree biomass, we employed the allometric equations of Kato et al. (Reference KATO, TADAKI and OGAWA1978) determined in the Pasoh Forest Reserve (2°60′N, 102°19′E), located 300 km south from the study site. These allometric equations are as follows (Figure 1):
$$\begin{equation}
{W_{{\rm{stem}}}} = 0.0313{\rm{ (}}{D_{{\rm{tree}}}}^2{H_{{\rm{tree}}}}{{\rm{)}}^{0.9733}}
\end{equation}$$\\
$$\begin{equation}
{W_{{\rm{branch}}}} = 0.136{W_{{\rm{stem}}}}^{1.07}
\end{equation}$$ \\
$$\begin{equation}
1/{W_{{\rm{leaf}}}} = 1/(0.124\,{W_{{\rm{stem}}}}^{0.794}) + 1/125,{\rm{ and}}
\end{equation}$$ \\
$$\begin{equation}
{A_{{\rm{tree}}}} = 11.4\,{W_{{\rm{leaf}}}}^{0.900},
\end{equation}$$
where W stem is the stem dry mass (kg), and A tree is the total leaf area of a tree (m2). Above-ground biomass of tree was thus W tree = W stem + W branch + W leaf.
Allometry comparison
We compared predicted Y values at the reference values of X among the two bamboo species and woody trees in the two ramet-scale and six genet-scale allometries from the perspective of light capture efficiency, i.e. H ramet (H tree) on D ramet (D tree) and W culm (W stem), W branch, A genet (A tree) and H genet (H tree) on W genet (W tree), and A genet (A tree), C genet (C tree) and CLAI on H genet (H tree). CLAI was calculated as A genet/C genet (A tree/C tree) (m2 m−2). To quantify the difference in predicted Y values between species, we chose three reference values of X (W genet, W tree, H genet or H tree), i.e. at the minimum and maximum points, and geometric midpoint across their range of overlap (Appendix 4).
Observation and estimation of biomass dynamics
By employing the architectural allometries, we estimated demographic turnover rates of stand-scale biomass of bamboos and trees. We first recorded the death and recruitment of bamboo ramets and tree stems and increment in stem diameter for trees for a subplot of 10 × 30 m for bamboos and 20 × 40 m for trees, in all the 24 plots (Appendix 3). The census intervals varied from 14 to 17 mo (first census was from February to July 2011, and the second census was from August to September 2012). For woody trees, we tagged stems with D tree ≥ 5 cm in the first census and recorded their survival, increase in diameter and recruitment to D tree ≥ 5 cm in the second census. We tagged and painted every culm of bamboo genets comprised of at least one ramet with D ramet ≥ 2 cm in the first census, and recorded dead and recruited ramets at the second census. To compare dynamics between bamboos and trees in the same light environment, we selected genets and trees located in the same layer of the forest canopy. Top height of trees (H tree) predicted by the allometry from stem diameter (D tree) at 5 cm was H tree = 6.7 m, i.e. the lower boundary height of tree census, and the corresponding genet sizes were n = 4 for G. ligulata (H genet = 6.7 m) and n = 3 for S. grande (H genet = 7.0 m). Therefore, genets with ramets equal to or more than these thresholds were used for comparison. Similarly, we applied an upper boundary D tree range to keep tree height within the height range of bamboo crowns. We employed D tree = 40.4 cm corresponding to H tree = 27.3 m, the maximum height observed in S. grande.
We estimated demographic turnover rates of above-ground biomass, based on the census records of D ramet (or D tree) and allometric equations that relate D ramet (or D tree) and biomass dimension, W ramet (or W tree). Above-ground biomass W stand (kg ha−1) was defined by W stand = [Σ jWj ]/Q, where Wj (kg) is biomass of the j-th individual genet (or tree) and Q (ha) is the area of a target plot (Appendix 3). To standardize time intervals, we employed the continuous-time model of the dynamics of above-ground biomass:
$$\begin{equation}
d{W_{{\rm{stand}}}}/dt = {R_{{\rm{stand}}}}{W_{{\rm{stand}}}} = ({B_{{\rm{stand}}}}-{M_{{\rm{stand}}}})\,{W_{{\rm{stand}}}},
\end{equation}$$
where R stand (y−1) is the net specific rate of biomass increase, B stand (y−1) is the instantaneous (not annualized) specific wood production rate, M stand (y−1) is the instantaneous specific wood loss rate and t is census interval (y) (Kohyama et al. in press). By integrating Eq. 6 with respect to the duration from t = 0 to T, we obtain estimates of R stand, M stand and B stand as (Phillips et al. Reference PHILLIPS, HALL, GENTRY, SAWYER and VÁSQUEZ1994),
$$\begin{eqnarray}
{R_{{\rm{stand}}}} & = & \big[ \ln ({W_{{\rm{stand, 0}}}}-{W_{{\rm{stand, death}}}} + {W_{{\rm{stand, growth}}}} \nonumber\\
&& +\, {W_{{\rm{stand, recruit}}}})-\ln ({W_{{\rm{stand, 0}}}}) \big]/T,
\end{eqnarray}$$
$$\begin{eqnarray}
{M_{{\rm{stand}}}} &=& \big[ \ln ({W_{{\rm{stand, 0}}}})-\ln ({W_{{\rm{stand, 0}}}} \nonumber\\
&&-\,{W_{{\rm{stand, death}}}}) \big]/T,{\rm{ and}}
\end{eqnarray}$$
$$\begin{eqnarray}
{B_{{\rm{stand}}}} &=& [\ln ({W_{{\rm{stand, 0}}}}-{W_{{\rm{stand, death}}}} + {W_{{\rm{stand, growth}}}}\nonumber\\
&&+\, {W_{{\rm{stand, recruit}}}}) {\rm{- ln}}\left( {{W_{{\rm{stand, 0}}}}{\rm{-}}{W_{{\rm{stand, death}}}}} \right){\rm{]/}}T{\rm{,}} \nonumber\\
\end{eqnarray}$$
where W stand, 0 (kg ha−1) is living biomass at the first census, W stand, death (kg ha−1) is a portion of W stand, 0 that was lost by death, W stand, growth (kg ha−1) is biomass increment due to tree growth during the census period (and is assumed zero for bamboo ramets), and W stand, recruit (kg ha−1) is the biomass increment due to newly recruited plants at the second census, and T (y) is census interval. Permanent plots in which a whole W stand, 0 was lost during two censuses were excluded from the following analysis. Likewise, plots in which W stand, 0 was zero were excluded from the analysis.
Response of biomass dynamics to canopy disturbance
We compared effects of the forest canopy conditions on biomass density, W stand, 0, and on the specific rate of biomass increase, R stand, for trees and bamboos as follows. Canopy conditions were evaluated by canopy LAI (m2 m−2) before logging in 2008 (LAIintact), the first census (LAI0), and by the loss of canopy LAI (LAIloss) between 2008 to 2011 (as logging took place in 2010). Canopy LAI was calculated for the trees with D tree ≥ 45 cm, which is the maximum size of trees protected from logging operations under the Selective Management System (SMS) of Malaysian forestry practice. This threshold of D tree for canopy trees was significantly larger than the subset of plants for biomass dynamics that was D tree ≤ 40.4 cm as stated above. We developed linear regression models to examine the effects of LAIintact on W stand, 0 and effects of LAI0 and LAIloss on R stand, separately for woody trees, G. ligulata and S. grande, with the model selection based on AICc (Hurvich & Tsai Reference HURVICH and TSAI1989). We carried out these analyses using R 3.3.3 (R Core Team, https://www.R-project.org/) with the package MuMIn for AICc calculation (https://CRAN.R-project.org/package=MuMIn) and raster and ggplot2 for mapping the study site (Appendix 1).
RESULTS
Allometric properties
The observed maximum ramet number per genet of G. ligulata and S. grande was 110 and 45, respectively. The mean and maximum diameters of all observed ramets were 3.7 cm and 7.6 cm for G. ligulata, and 5.5 cm and 10.3 cm for S. grande, respectively. The maximum crown height among all observed genets was 17.5 m for G. ligulata and 27.3 m for S. grande. The maximum stem diameter at breast height of woody trees was 182 cm and the maximum height was 49.4 m. The estimated parameters of the allometric equations used are given in Table 2 (bamboos) and Table 3 (trees).
Table 2. Estimated allometric parameters for two bamboo species: Gl for Gigantochloa ligulata and Sg for Schizostachyum grande. The estimated scaling exponents (α) and allometric constants (β) between two dimensions X and Y, Y = α ln X + β, are shown for 95% confidence interval. Refer to Figure 1 for the form of equations. R 2 is the coefficient of determination of the models. Type 1 represents two bamboo species shared common α and β; Type 2 represents difference between the two species only in β; and Type 3 represents difference between the two species in α (therefore, also in β). Parameters identical between the two species are showed only for G. ligulata. P-values of all the estimated α were less than 0.001.
Table 3. Estimated allometric parameters for trees. N represents the number of samples. The mean estimated coefficients of α, β, and γ along with confidence intervals are given for equations estimating height (H tree) and canopy area (C tree) as a function of stem diameter (D tree): 1/Y = 1/(βXα ) + 1/γ. R 2 is the coefficient of determination of the models. P-values of all the estimated parameters were less than 0.001.
Compared with tree stems, both ramets and genets of the two bamboo species had more efficient crown architectures in terms of biomass expenditure for displaying leaves. Culms of S. grande attained 1.3 times more height than woody tree stems at a diameter of 6.6 cm (Figure 2a). This value of 6.6 cm corresponds to the geometric midpoint of the range of the observed diameter distribution (Appendix 4). On the basis of culm (stem) biomass, there were similar contrasts between the top height of bamboo culms and tree stems, where both G. ligulata and S. grande culms were 1.5 times as tall (Figure 2b). The bamboo culms were on average taller than trees, even though some of their culms were slanted or bent (Appendix 2), which would also contribute to the large variation in height at a given culm diameter or biomass. At the genet-scale comparison, G. ligulata and S. grande allocated 2.7 and 1.8 times more in lateral branch biomass, and supported 4.4 and 4.2 times larger leaf area, than trees, respectively (Figure 3a-b), at an above-ground biomass of 30 kg for G. ligulata genets vs. trees and an above-ground biomass of 50 kg for S. grande genets vs. trees. With respect to crown height at the geometric midpoints of above-ground biomass, trees on average had an intermediate height between G. ligulata and S. grande. These bamboos attained heights of 0.79 and 1.1 times that of trees, respectively (Figure 3c).
Figure 2. Allometric properties of bamboo ramets and woody trees in a Malaysian hill rain forest. Ramet (stem) top height (H ramet, H tree) on culm (stem) diameter at breast height (D culm, D tree) (a) and on culm (stem) biomass (W culm, W stem) (b) for Gigantochloa ligulata (orange), Schizostachyum grande (pink) and woody trees (grey). Solid lines show the allometric predictions.
Figure 3. Allometric properties relative to above-ground biomass (W genet, W tree) for genets and stems in a Malaysian hill rain forest; Gigantochloa ligulata (orange), Schizostachyum grande (pink) and woody trees (grey). Solid lines show the allometric predictions. Allometric relation with branch biomass (W branch) (a), leaf area (A genet, A tree) and crown top height (H genet, H tree) (c).
At reference crown heights of 7.7 m for G. ligulata and 9.7 m for S. grande, bamboos displayed 5.9 and 3.0 times more total leaf area, and 6.6 and 6.7 times larger crown projection area than trees of the same crown height, respectively (Figure 5a-b). As a consequence of more leaves in the larger crown area, G. ligulata and S. grande constructed crowns of similar or lower density at reference heights, resulting in crown leaf area indices that were 10% and 55% lower than trees, respectively (Figure 5c).
Between the two bamboo species, ramet-scale allometry of top height on culm diameter and that on culm biomass were not significantly different (Figure 2a-b, Table 2). At the genet-scale, G. ligulata allocated 1.4 times as much above-ground biomass to lateral branches as S. grande (Figure 3a). Accordingly, the height of the genet crown of G. ligulata was 20% shorter than that of S. grande at 14 kg above-ground biomass (Figure 3c), which was partly attributable to the fact that G. ligulata genets consisted of a larger number of smaller ramets as compared with a S. grande genet at the same above-ground biomass (Figure 4). Despite the difference in biomass allocation to lateral branches between the two bamboos, total leaf area at the same above-ground biomass was similar (1.1 times as large in G. ligulata as in S. grande) (Figure 3b). At the geometric midpoint of the measured crown height range for both bamboos (7.1 m), G. ligulata displayed 1.7 times as much leaf area as did S. grande (Figure 5a), while crown area of the two bamboos did not show a large difference (10% larger in G. ligulata) (Figure 5b). Therefore, CLAI of G. ligulata was 1.5 times that of S. grande at the reference height, although variation in predicted CLAI was large (Figure 5c).
Figure 4. Distribution of culm diameter (D ramet) in genets of Gigantochloa ligulata (a, b) and Schizostachyum grande (c, d) in a Malaysian hill rain-forest. Left panels (a, c): average D ramet per genet (D genet) vs. number of ramets per genet (n). Right panels (b, d): change in D ramet distribution per genet along the genet rank in D genet. The lower and upper limits of the box plot in right panels represent 25- and 75-percentile D ramet, respectively, and midpoint bar of each box shows D genet. Whiskers represent the minimum and the maximum D ramet. Inset numbers indicate n.
Figure 5. Allometric properties relative to top crown height (H genet, H tree) for genets and stems in a Malaysian hill rain forest: Gigantochloa ligulata (orange), Schizostachyum grande (pink) and woody trees (grey). Solid lines show the allometric predictions. Allometric relation with leaf area (A genet, A tree) (a), crown area (C genet, C tree) (b), and crown LAI (CLAI) (c).
Population and biomass dynamics
In the 2011 census of 24 permanent plots after the logging experiment, there were 2023 ramets and 132 genets of G. ligulata with the number of ramets n ≥ 4, and 982 ramets and 130 genets of S. grande with n ≥ 3. In these subplots, there were 1508 stems of trees including 67 canopy trees (stem diameter D tree ≥ 45 cm) and 1424 stems of middle-canopy trees sharing the subcanopy layer with bamboos (D tree ≤ 40.4 cm).
Above-ground biomass of trees was 341 Mg ha−1, and that for G. ligulata and S. grande was 3.3 and 4.0 Mg ha−1, respectively. In contrast to the 100-fold difference in biomass, stand-level LAI was 4.38, 0.68 and 0.52, for trees, G. ligulata and S. grande, respectively. Demographic turnover rate of biomass for bamboo ramets was 4.5 times or more that of woody trees in the subcanopy layer (Figure 6, Appendix 5). The average of estimated instantaneous specific wood loss rate (M stand) was 0.16, 0.22 and 0.033 y−1 for G. ligulata, S. grande and trees, respectively, whereas instantaneous specific wood production rate (B stand) was 0.27, 0.43 and 0.057 y−1 for G. ligulata, S. grande and trees, respectively (Appendix 5). Therefore, average rate of net specific biomass increase (R stand = B stand – M stand) was +0.11, +0.27 and +0.024 y−1 for G. ligulata, S. grande and trees, respectively (Appendix 5).
Figure 6. Demographic dynamic rates of above-ground biomass for Gigantochloa ligulata (orange), Schizostachyum grande (pink) and woody trees (grey) in a Malaysian hill rain forest (a); symbol size represents the degree of LAI loss by timber logging (LAIloss) at each study plot. The relation between LAIloss and net specific rate of biomass increase (coarse wood production rate relative to above-ground biomass), R stand (b). Solid lines show predictions by selected empirical models in Table 4, where symbol size represents the LAI at the first census (LAI0).
The second census of the 24 subplots showed decreasing mean stand LAI from pre-logging conditions in 2008 of 1.70 m2 m−2 (LAIintact) to post-logging conditions in 2011 of 1.42 m2 m−2 (LAI0), where lost LAI due to logging (LAIloss) varied from 0 to 3.76 m2 m−2 among the plots (Appendix 6). Stand biomass (W stand, 0) for G. ligulata showed a negative correlation with LAIintact (R 2 = 0.313), while that of S. grande and subcanopy trees had no correlation with LAIintact (Table 4). Net specific rate of biomass increase (R stand) of G. ligulata during the census period showed a positive correlation with LAIloss (R 2 = 0.548), while that of S. grande did not correlate with either LAI0 or LAIloss (Figure 6). In contrast to G. ligulata, R stand of subcanopy trees was negatively correlated with LAIloss (Figure 6) (R 2 = 0.251).
Table 4. Coefficients of the best models from the analyses of the effects of canopy structure on W stand, 0 and R stand. Refer to Appendix 5 for the focal sizes of D tree and n for response variables. R 2 is the coefficient of determination of the models.
DISCUSSION
To understand why tropical bamboos persist in mature forest stands and recover vigorously after disturbance, we proposed the hypothesis that the architecture of bamboos is more efficient for light capture than that of woody trees, and tested this hypothesis by examining the architectural and dynamic properties of bamboos in comparison with trees. Observation of the two most abundant bamboo species, Gigantochloa ligulata and Schizostachyum grande, revealed that, as compared with woody trees, hollow culms have greater height attainment per ramet and greater total leaf area per genet, as the less costly hollow stems allowed for higher biomass allocation to branches resulting in a larger crown projection area. Bamboos exhibited higher biomass turnover rates of ramet culms than tree stems. Taken together these findings suggest how bamboos respond to environmental change. This inference was also supported by our results that G. ligulata increased biomass in response to canopy removal by disturbance.
In our study, the specific rate of biomass increase to above-ground biomass was based on recruitment rates of ramet culms and tree stems, and radial growth rate of tree stems (but not for bamboos). To fully evaluate net primary production rate (NPP) of above-ground biomass, we further need to account for the fine-litter production rate (Clark et al. Reference CLARK, BROWN, KICKLIGHTER, CHAMBERS, THOMLINSON, NI and HOLLAND2001). The rate of fine-litter production is defined as the rate of branch and leaf fall from living plants. Clark et al. (Reference CLARK, BROWN, KICKLIGHTER, CHAMBERS, THOMLINSON, NI and HOLLAND2001) compiled the ratio of annual rates of litterfall to above-ground biomass in 39 tropical forests (average 0.032 ± 0.021 y−1). Specifically, for the Pasoh Forest Reserve in Peninsular Malaysia they found the ratio of 0.025 y−1 (see also Kira Reference KIRA, Tomlinson and Zimmermann1978). Therefore, the specific biomass increment rate of bamboos (0.27 and 0.43 y−1 for G. ligulata and S. grande, respectively) is three to five times higher than the estimate for trees 5.0–40.4-cm in dbh (0.082 y−1, Appendix 5). The difference in above-ground NPP between bamboos and trees would be even larger if the leaf fall from bamboo ramets was taken into account. Moreover, by including trees in the canopy layer (dbh > 40.4 cm), which consist of the dominant portion of above-ground biomass, specific NPP to above-ground biomass would be far smaller for trees than for bamboos.
Despite the marked difference in architecture and dynamics between bamboos and woody trees, the key features relevant to light capture were similar, such as crown height and crown LAI at the genet scale (Figure 3c, 5c). In the case of woody trees, constraints of mechanics and hydraulic requirements of stems and branches limit the amount of foliage and crown height that can be sustained by a stem of specified basal area (Niklas Reference NIKLAS2007). A bamboo ramet is composed of a culm and branches with hollow interior and thus has a large sapwood area per unit basal area (McClure Reference MCCLURE1966, Wong Reference WONG1995), which enables them to maintain substantial amount of foliage per unit above-ground biomass. Furthermore, the hollow interior efficiently reduces culm biomass without compromising physical strength, and this allows bamboo ramets to allocate larger biomass to lateral branches and attain 1.5 times the height of woody trees at the same above-ground biomass. Therefore, the separation of biomass into numerous culms within a genet, although it does not contribute to height gain, is compensated for by the cheap ramet architecture for bamboos, and which results in similar above-ground biomass for height gain between bamboo genets and woody trees.
Even on the basis of culm/stem diameter, the top height of ramets, in particular of S. grande, was still higher than woody trees, though this is not readily explained by the efficiency of hollow culms. From a wood mechanical perspective, loadable weight and height that meet buckling safety limits should decrease with decreasing stem diameter (King et al. Reference KING, DAVIES, TAN and NUR SUPARDI2009, McMahon Reference MCMAHON1973). The observed higher allocation to lateral branches on thin culms should thus increase the risk of mechanical failure. However, the clumping genet structure of these bamboos, where adjacent ramet crowns may press against each other or intertwine their branches, could enable them to exceed the stability limits for a single stem. In addition, the extreme slenderness of the culms allows them to bend their tops to the ground without breaking, as is the case for slender-stemmed shrubs (Larjavaara Reference LARJAVAARA2015). Outward bending by ramets could benefit the genet by increasing its crown area and shading adjacent saplings of trees. Furthermore, the quick replacement of short-lived ramets should reduce the deleterious effects of any breakage due to mechanical instability and/or falling tree branches.
Bamboo genets successfully expanded their crowns by a larger amount of branches (Figure 3b), which is an architectural strategy that places higher importance of the light capture at a current light availability, instead of the investment in height gain to get better light availability (King Reference KING1990, Kohyama Reference KOHYAMA1987, Kohyama & Hotta Reference KOHYAMA and HOTTA1990, Yamada et al. Reference YAMADA, YAMAKURA and LEE2000). As a consequence, the bamboos exhibited crown LAI in a similar range to that of woody trees, indicating that from the perspective of the architecture the two bamboo species possess shade tolerance to the extent of the average of the woody tree community.
Although bamboos exhibited efficient architectures, this does not assure their persistence as a species in tropical rain forest. The mechanism of plant species coexistence in forests has been partly explained by the separation of successional niches under a horizontally heterogeneous canopy structure (Rees et al. Reference REES, CONDIT, CRAWLEY, PACALA and TILMAN2001). Along the spectrum of pioneer to climax species, bamboos might appear to be pioneers, as they exhibit quick recovery after forest disturbance. However, a number of typical properties of pioneer species (Swaine & Whitmore Reference SWAINE and WHITMORE1988) are not congruent with that of bamboos, and thus, the mechanism of bamboo persistence in the rain forest is not readily obvious in the context of successional niche theory. Kohyama & Takada (Reference KOHYAMA and TAKADA2009, Reference KOHYAMA and TAKADA2012) developed another theory of species coexistence under the presence of one-sided competition for light along vertical foliage distribution, where demographic advantages permit the persistence of lower-canopy species without the aid of gap dynamics. As a clonal plant, which is known to be robust in species competition (Svensson et al. Reference SVENSSON, RYDIN, CARLSSON and van der Maarel2005, van Groenendael et al. Reference VAN GROENENDAEL, KLIMEŠ, KLIMEŠOVÁ and HENDRIKS1996), shading (van Groenendael et al. Reference VAN GROENENDAEL, KLIMEŠ, KLIMEŠOVÁ and HENDRIKS1996) and resource fluctuations (Suzuki & Stuefer Reference SUZUKI and STUEFER1999), a genet of bamboos is expected to maintain lower genet mortality under a forest canopy. As indicated by the wide crown architecture, bamboos also possess a certain degree of shade tolerance. These demographic advantages of bamboos fulfil conditions required for the coexistence with taller tree species in vertically stratified forest canopy (Kohyama & Takada Reference KOHYAMA and TAKADA2009). Furthermore, with their unique crown structure, bamboos possibly differentiate the vertical distribution of foliage in a life-form-specific manner distinguished from that of co-occurring tree species, and this provides a possibility of coexistence with similar-statured species as predicted by foliage partitioning theory (Kohyama & Takada Reference KOHYAMA and TAKADA2012).
To understand how bamboos persist in a wide range of light conditions, previous studies suggested the importance of high photosynthetic plasticity of bamboo leaves in response to light (Montti et al. Reference MONTTI, VILLAGRA, CAMPANELLO, GATTI and GOLDSTEIN2014, Yang et al. Reference YANG, SUN, ZHANG, COCHARD and CAO2014). The present study sheds new insights on the uniqueness of bamboo architecture and its biomass dynamics. Instead of having large single stems like subcanopy tree species, the life-form of bamboos, which descended as a subfamily of Poaceae, is characterized by that of typical perennial grasses, with modular architecture, absence of cambial growth, and large subsidies of ramet recruitment from the rest of genet. Trees typically change their architecture by additions to their existing crowns, which are limited by relatively slow cambial growth in stems and branches. Hence they have a limited capacity to respond to rapid environmental change. This explains our observation that trees showed a decline in coarse wood production rate in disturbed sites with high loss of LAI. The combination of the grass life-form and woody culms with a hollow interior permits bamboos to quickly replace their cheap ramets while maintaining a similar crown architecture of genets as woody trees. In contrast to the light capture strategy of dicotyledonous trees, the dynamic maintenance of crown architecture is a feat attributable to the swift response to canopy disturbance by many bamboo species (Gagnon et al. Reference GAGNON, PLATT and MOSER2007, Soderstrom & Calderon Reference SODERSTROM and CALDERON1979, Zhang & Cao Reference ZHANG and CAO1995).
Such advantages of bamboos are, however, likely less significant in primary lowland tropical rain forests with extra-tall canopy structure and limited disturbance events, where the high turnover nature of ramet populations becomes costly. As King (Reference KING1991) reported, saplings of shade-tolerant tree species could allocate more than 60% of above-ground biomass production to foliage in deep shade, while reusing the supporting platforms of stem and branches, which contrasts with the bamboo life-form. In this case of more pervasive shade in forest understoreys (Tang et al. Reference TANG, KACHI, FURUKAWA and AWANG1999), one-sided competition may facilitate the coexistence of shade-tolerant trees, rather than bamboos. In fact, throughout humid lowland areas of South-East Asia, bamboos are rare outside of disturbance-prone environments, such as logged forests, steep slopes and riparian stands. Nonetheless, the high abundance of our study species down to 600 m asl suggests that bamboos are an important component across the widespread uplands of South-East Asia (Dransfield Reference DRANSFIELD1992, Wong Reference WONG1995).
ACKNOWLEDGEMENTS
We extend our sincere thanks to Perak Integrated Timber Complex, Forest Department Peninsular Malaysia Headquarters and State Forestry Departments of Peninsular Malaysia for permission and support of fieldwork in the Temenggor Forest Reserve. We are grateful to the staff of the herbarium of Forest Research Institute Malaysia for invaluable support and advice on field study. We are also grateful to Dr Tatsuyuki Seino for his advice in the field. This study was supported by Grants-in-Aid for Scientific Research from the Japan Society for the Promotion of Science (Nos. 21405006), Global COE program of Hokkaido University, the Global Environment Facility through the United Nations Development Programme Malaysia (MAL/04/G31) and the International Tropical Timber Organization (PD 16502 Rev.3 [F]). In-kind support was provided by the Government of Malaysia through the Ministry of Natural Resources and Environment and Forest Research Institute Malaysia. We also thank D. King and one anonymous reviewer for their valuable and detailed comments, which greatly improved this paper.
Appendix 1. Map of study site with 20-m contour lines. The thick border line traces the boundary of the study site (200 ha). Square symbols represent the location of 24 permanent plots, among which open squares indicate plots where tree crown architecture was observed. Circle and triangle symbols represent locations of Gigantochloa ligulata and Schizostachyum grande genets sampled for allometric analysis, respectively. For the destructive sampling, we searched genets in target size classes along logging trails by paying approximately equal efforts to the southern and northern slopes of the study site. Canopy condition above sampled genets were classified into three categories optically: closed canopy (C), intermediate (I) and open canopy (O). The numbers of sampled genets at each canopy condition (C, I and P) were 11, 9 and 9 for G. ligulata, and 9, 8 and 8 for S. grande, respectively.
Appendix 2. Scheme of architectural observation and estimation for a bamboo ramet and genet. Above-ground biomass and total leaf area of genets, W genet and A genet, were estimated by summing the ramet-scale dimensions which were obtained applying ramet allometries for each D ramet (refer to Figure 1 and Table 2).
Appendix 3. Scheme of sampling area in permanent plots. The whole 20 × 80-m plot in grey was used to collect data on crown architecture of trees (D tree, H tree and C tree). The central 20 × 40-m area in solid black was used to collect data on demography of bamboo ramets and tree biomass dynamics. The central 10 × 30-m area shown within the dashed line was used to collect data on genet architecture (Appendix 2) and biomass dynamics of bamboos.
Appendix 4. Allometric comparisons at three reference points on X value. First from an overlapped range between two comparing species, the minimum and the maximum values of the range were determined (X min and X max, respectively), and then, the geometric midpoint, which is the midpoint between log(X min) and log(X max), was determined (X mid). Predicted Y values on reference points were compared between two species: species 1 vs. species 2 (a); species 1 vs. species 3 (b); and species 2 vs. species 3 (c).
Appendix 5. Mean of biomass dynamic rates (y−1). m, wood loss rate; b, wood production rate; r, relative increment rate; N, the number of stems or culms used for the calculation, where the value in parentheses after N represents the number of culms used for b estimation; SD for standard deviation. The value in parentheses after b is estimation including fine-litter production rate, 0.025 y−1, obtained by Clark et al. (Reference CLARK, BROWN, KICKLIGHTER, CHAMBERS, THOMLINSON, NI and HOLLAND2001) in the Pasoh Forest Reserve in Peninsular Malaysia.
Appendix 6. Variation of LAI and stem basal area (BA) of woody trees among permanent plots before (LAIintact, BAintact) and after (LAI0, BA0) timber logging, and change (LAIloss, BAloss) due to logging operation. Column D tree represents lower size limit of trees for calculation. Columns min, average, max, and SD show minimum, average, maximum and standard deviation among 24 permanent plots.
