Trade‐off between standing biomass and productivity in species‐rich tropical forest: Evidence, explanations and implications

Despite its broad implications for community structure and dynamics, we lack a clear understanding of how forest productivity is partitioned among tree species. As leaf mass per unit of standing biomass declines with tree size, species achieving larger stature should show lower relative productivity as compared to smaller stature species. However, many observations indicate large‐stature species grow faster than small‐stature species. In this study, we address this apparent paradox, and clarify interspecific trade‐offs between turnover rates and maximum size in terms of forest‐level productivity and biomass storage. We examined data from the 1990 and 2000 surveys of the Pasoh 50‐ha plot of Malaysian rain forest. In these data, individual stems ≥1 cm stem diameter (dbh) have been identified, marked, measured and mapped. We applied site‐specific equations to estimate tree biomass from dbh. We estimated species‐level productivity and loss rates that are less influenced by census interval‐related effects and biases. Among 390 abundant tree species, species with high stand‐level biomass were predominantly those large‐stature species where individuals could achieve large sizes. We found that relative (= per‐species‐biomass) productivity and loss rate, per‐capita recruitment and mortality of species were all negatively correlated to species biomass and maximum size, but not to species abundance. Large‐stature species grew faster than small‐stature species at the same tree sizes up to 36 cm dbh. However, the relative growth of large species at their maximum size markedly declined. As a result, tree‐level relative growth at maximum size and species‐level relative productivity decreased with species‐level biomass. Performing further analyses using smaller plots in four old‐growth forests in Indonesia and Japan, we observed a similar interspecific negative relationship between relative productivity and biomass. We expect this phenomenon is widespread across species‐rich forests. Synthesis. How productivity is partitioned among species determines and reflects forest ecosystem functioning and species coexistence. Many species with low biomass and small maximum stem sizes disproportionally contribute to forest primary productivity, rapid recovery and resilience. In contrast, long‐lived, high‐biomass species contribute disproportionally to ecosystem stability and carbon storage. This complementarity reflects differentiation by adult stature among species.


| INTRODUC TI ON
The diversity of life histories found within mixed species communities influences coexistence and ecosystem properties (Falster, Brännström, Westoby, & Dieckmann, 2017;Hooper et al., 2005;Liang, Zhou, Tobin, McGuire, & Reich, 2015). Previous studies have found that net primary productivity is correlated with community species diversity at the level of forest stands (Chisholm et al., 2013;Jucker et al., 2016;Mori, 2018). However, it remains unclear how stand-level productivity is partitioned among co-occurring tree species with differing life histories.
Past studies of stand productivity have various shortcomings.
Most have neglected spatial variation and differences among species (Kira & Shidei, 1967;Malhi et al., 2004;Phillips et al., 1998). When studies have examined interspecific variation, they have typically neglected stand-level processes and focused on selected, for example, common species (Condit, Sukumar, Hubbell, & Foster, 1998;Kohyama, Suzuki, Partomihardjo, Yamada, & Kubo, 2003;Poorter et al., 2008). Furthermore, conventional estimates of net primary production by tree growth (Clark et al., 2001;Kira & Shidei, 1967;Ohtsuka et al., 2005)-biomass gained through growth of surviving stems and by new recruits divided by the census interval-are problematic for a number of reasons including their inability to account for productivity during inter-census intervals (Malhi et al., 2004;Talbot et al., 2014). To remedy these shortcomings, we developed a method for estimating instantaneous rates of production and loss that is less impacted by census interval effects and biases (see Kohyama, Kohyama, & Sheil, 2019). In addition, by defining stand-level production as the sum of species-level production, biases due to ignoring interspecific heterogeneity are avoided. Using this new approach, we clarify how forest production is partitioned among tree species.
If the largest tree sizes in a population represent the populationlevel biomass, then a pattern similar to stand-level biomass distributions should arise (Bastin et al., 2018;Lutz et al., 2018;Slik et al., 2013): large-stature species have higher species biomass as compared to small-stature species. We further expected that the relative growth and mortality of individual trees at the species maximum size would represent species-level relative productivity and loss, respectively ( Figure 1). Therefore, it is necessary to relate species-level biomass turnover to species demography as measured by tree-size dependent growth and mortality.
Species-level biomass turnover (i.e. per-biomass relative rate of production and loss) is either negatively, positively or not related to species biomass in a community. The negative turnover-biomass correlation could be expected from the following assumptions. The relative growth of individual forest trees generally decreases with increasing size in any species population (Iida et al., 2014;Kohyama, Potts, Kohyama, Abd Rahman, & Ashton, 2015). This trend likely 5. Performing further analyses using smaller plots in four old-growth forests in Indonesia and Japan, we observed a similar interspecific negative relationship between relative productivity and biomass. We expect this phenomenon is widespread across species-rich forests.
6. Synthesis. How productivity is partitioned among species determines and reflects forest ecosystem functioning and species coexistence. Many species with low biomass and small maximum stem sizes disproportionally contribute to forest primary productivity, rapid recovery and resilience. In contrast, long-lived, high-biomass species contribute disproportionally to ecosystem stability and carbon storage. This complementarity reflects differentiation by adult stature among species.

K E Y W O R D S
adult stature, biodiversity, demography, ecosystem function, growth, mortality, plant population and community dynamics, species coexistence F I G U R E 1 Assumed link between species-level biomass productivity, and size-dependent tree growth trajectory. Circles indicate species maximum individual tree values for large-, intermediate-and small-stature species. (a) Forest-wide decline of relative growth with tree size predicts a negative relationship between relative productivity and biomass. (b) Interspecific differentiation in growth curves may bring about an increasing relationship between relative productivity and biomass reflects the reduced ratio of photosynthetic leaves, and thus available energy capture, in relation to total tree biomass (Enquist & Niklas, 2002;Poorter et al., 2012Poorter et al., , 2015. Assuming that the decline of relative growth with tree size is similar among species with varied maximum sizes then large-stature species should show lower relative growth at maximum tree size and species-level relative productivity ( Figure 1a). Tree mortality at maximum size and species-level relative loss would also be lower in large-stature species than in smallstature species (Iida et al., 2014;Kohyama et al., 2015). However, at the stand-scale, larger (and taller) trees capture a greater share of available light than smaller (and shorter) trees, which may compensate for the cost of achieving large tree size.
In reality, there is interspecific variation in size-dependent growth of trees. Observations in tropical forests suggest that species with larger stature typically grow faster, have lower mortality and recruit less frequently as compared to species with smaller stature (Iida et al., 2014;King, Davies, & Noor, 2006;King, Wright, & Connell, 2006;Kohyama et al., 2003Kohyama et al., , 2015Lieberman, Lieberman, Hartshorn, & Peralta, 1985;Manokaran & Kochummen, 1987;Poorter et al., 2008;Rüger et al., 2018). Smaller stature species compensate for slow growth by maturing sooner and recruiting more frequently. Theoretically, lower mortality-to-growth ratios over a tree size range results in larger maximum tree sizes in stable populations (Kohyama et al., 2015). Sillett et al. (2010) andStephenson et al. (2014) suggest that absolute (not relative) productivity of forest trees steadily increases with tree size; though the evidence remains ambiguous (Ligot et al., 2018;Sheil et al., 2017). In addition, mortality does not simply decline with tree size, but shows a minimum among mid-sized trees, with increasing rates of both smaller and larger stems (Coomes & Allen, 2007;Iida et al., 2014;King, Davies, et al., 2006;King, Wright, et al., 2006;Kohyama et al., 2015;Rüger, Huth, Hubbell, & Condit, 2011). We wish to examine how tree mortality at species maximum size relates to total species biomass. Therefore, it remains possible that species-level biomass turnover (relative productivity and loss) is positively related to species biomass ( Figure 1b).
To guide our work, we propose the following four complementary hypotheses: (1) In a mixed-species stand, species that achieve large sizes (i.e. in which some individuals achieve large diameters and stature) have higher species-population biomass than smaller sized species (n.b. this is not a truism: it would be false if small-stature species were sufficiently common and large-stature species sufficiently rare), (2) species population-wide relative production rates are lower for higher biomass species; (3) low species-level relative productivity for high-biomass species is explained by large-stature species possessing a lower proportion of photosynthetic leaves to wood mass than small-stature species and (4) despite the fact that individual trees of large sized species grow faster than those of small sized species at equivalent individual stem sizes, marked reduction of tree-level relative growth at large tree sizes brings about lower species-level relative productivity for large-stature, high-biomass species.
To test these hypotheses, we examined data from the Pasoh 50-ha forest plot in Peninsular Malaysia. We estimated and compared the biomass turnover (productivity and loss) and abundance turnover (recruitment and mortality) of 390 co-occurring tree species. From repeated tree inventory data, we generated 'identity-free' data, in which we replaced the species identity and spatial position of a stem with another stem of similar diameter. Comparing results of observed and identity-free communities allowed us to disentangle species properties from effects of forest-wide tree size dependence.
We also quantified stem size dependence of individual growth and mortality for every species, to relate biomass turnover by ontogeny and life history. We examined the generality of our results by analysing data from four other old-growth forests.

| Pasoh plot data
We used data from the 50-ha forest dynamics plot (2°59′N,  (Condit et al., 1999;Davies, Noor, LaFrankie, & Ashton, 2003;Manokaran & LaFrankie, 1990). We used data from two censuses (~1990 and ~2000). For individual trees, the time interval varied from 10.02 to 11.25 years. Biomass estimates were determined using allometric equations (Niiyama et al., 2010) calibrated using destructive sampling at Pasoh in 2004-2005; these equations estimate tree height (m), leaf mass and stem plus branch mass, and coarse root mass (Mg oven dry mass) from dbh (cm). We estimated leaf mass and above-ground mass (i.e. leaf mass plus stembranch mass) of every tree in two censuses using these in situ equations. These equations disregard interspecific differences though we note that the allometries of tree height versus dbh in the Pasoh plot are not significantly different among the most abundant 200 tree species, and across maximum species sizes (Iida et al., 2011). We examined the effect of interspecific variation in stem wood density using the Global Wood Density Database (Zanne et al., 2009). Large declines in measured dbh between censuses, where the change in log dbh between the two censuses was smaller than -0.1 (or more than c. 10% reduction), were considered as the death of one stem and recruitment of a new stem by regrowth (Kohyama et al., 2015).
To disentangle the effects of species-specific demographic properties and response to local site conditions from effects of forest-wide average tree-size dependence, we generated an 'identityfree' dataset, in which demography is independent of species identity and location ( Figure 2). To generate the identity-free data, we first sorted the trees in the observed data in order of their dbh in the 1990 census (tied values were randomized), and swapped species identity and location between every neighbouring pair of odd-and even-number trees in the sorted data, leaving other tree measures (dbh in two censuses and time interval) unchanged. By using this procedure, each identity-free 'species' population possesses a treesize distribution similar to that observed, but growth, survival and recruitment are independent of species identity and site (e.g. topography, soil, neighbours and shade).
We examined variation in turnovers for biomass and tree density among species populations in the entire 50-ha plot and in two hundred 50-by-50 m 0.25-ha subplots, for both observed and identity-free data. We selected species populations with 100 or more surviving trees (≥1 cm dbh) in the 50-ha plot, and those ≥20 trees in each 0.25 ha subplot. We also generated an aggregated population consisting of all unselected species, but included it only when we estimated forest-level turnover.
We quantified turnover in biomass B (Mg/ha) and in tree density or abundance N (ha -1 ) for each species population as follows. We denoted the individual-tree above-ground oven-dry biomass (Mg) of a tree i in the first and second census by W 0i and W Ti respectively. If a tree i was recorded in the first census but then died before the second, we set W Ti = 0; else if a tree i recruited and was recorded only in the second census, W 0i = 0. We denoted the state of survival s i , and that of death d i , such that s i = 1 only if a tree i was alive in both censuses (otherwise s i = 0), and that d i = 1 only if i died during censuses. The third state, that a tree i recruited, is thus 1 − s i − d i . Above-ground biomass of a species population at the first and second census are B 0 = Σ i W 0i /A and B T = Σ i W Ti /A (Mg/ha) for all i's, respectively, and the first census biomass for trees that survived until second census is where A (ha) is the horizontal land area of plot (or subplot). Tree density, or abundance of species in the two censuses are , respectively, and that survived over two We estimated the instantaneous production and loss of biomass B of each species based on a 'continuous-time' model, where p (year -1 ) is relative (= per-biomass) production, or relative productivity, by tree growth including recruits' ingrowth, and l (year -1 ) is relative loss by tree mortality (see Kohyama et al., 2019). For the dynamics of abundance N, we employed the same formulation as Equation 1: where r and m are instantaneous per-capita recruitment and mortality, respectively . If the time interval between two censuses is identical at T (year) across all trees, we obtain turnover estimates as: p = ln(B T /Bs 0 )/T, l = ln(B 0 /Bs 0 )/T, r = ln(N T /Ns)/T and m = ln(N 0 /Ns)/T (Kohyama et al., , 2019.
However, because census duration varies among trees in the Pasoh plot data, we employed the following implicit equations of turnover obtained by time integration of Equations 1 and 2 from t = 0 to T i for every tree i, We solved Equations 3-6 using the Newton-Raphson method (cf. Kubo, Kohyama, Potts, & Ashton, 2000). We note that these descriptive estimates of turnover are prone to estimation errors. Per-capita vital rate estimates (r, m) are influenced by the count of states, that is, survival, death and recruitment  and per-biomass rates (p, l) are influenced by those events as well as measurement errors of dbh. In the next section we introduce statistical models to quantify dbh-dependent growth and mortality of each species population by the combination of community-wide model parameters and species-specific deviation from those (as random effect). We compare (p, l) and these statistical estimates of growth and mortality at the largest population tree size.
We quantified period-mean population biomass B (and so for leaf biomass, denoted B L ) and tree density N during the period over two censuses (Kohyama et al., , 2019 as In addition to B and N, we quantified each species population by the 99 percentile tree biomass W max (Mg) as a standardized maximum tree size to avoid the effect of population size N (Kohyama et al., 2015).
Plot-level biomass B plot is Σ k B, where k denotes species (including the aggregated population of all rarer species). Plot-level absolute F I G U R E 2 Generation of identity-free data from measurements by replacing tree demography between each pair of trees with the most similar stem diameters. The two squares represent two distinct local sites from within the entire plot, where each of the paired trees is at the centre. Circles show trees, and the size of circles indicates initial tree diameter. In the identity-free data, tree diameters in two censuses are replaced between the pair of trees (white vs. black circle). This replacement was performed for every other pair of trees so that all trees were swapped once Replace tree demography production by tree growth and recruitment, P plot and absolute loss by tree death L plot are: (Kohyama et al., 2019). Niiyama, Ripin, Yasuda, Sato, and Shari (2019) compiled the records of monthly rate of fine litter fall, that is, leaves, reproductive organs, twigs and bark, etc., in 100 traps of 50 m 2 in total, over 25 years in the Pasoh forest. We employed the records of the 8 years from June 1992 to May 2000, the period overlapped with the examined two tree censuses. By denoting the absolute rate of fine litter fall F (Mg ha -1 year -1 ), and assuming no loss between production and fall of these fine parts, we obtained an estimate of above-ground net primary production NPP to be:

| Stem-size dependence of individual growth and mortality
We used the plot-wide Bayesian procedure (Kohyama et al., 2015) to estimate tree-size-dependent growth for each tree of species. We fitted a curve relating relative rate of biomass increase of a surviv- Equations 6 and 7), where D = f(W) is the reverse function of the set of allometric functions by Niiyama et al. (2010). We determined a nearly perfect approximation f(W) = 22.4W 0.326 exp(0.283W 0.203 ). We approximated among-tree variation in g(W) at any given W by exponential distribution, and accounted for dbh measurement error. Similarly, we fitted a curve relating tree mor- We set all (α, β, γ)'s of g(W) and μ(W) for each species to the plot-wide value and a species-specific parameter. Treatment of measurement error and model-parameter priors follows those in Kohyama et al. (2015).

| Other plot data
We performed similar analyses using data from four other old-growth mixed forests. Serimbu plots (two plots, 1 ha each and combined in the analysis) are located in an intact lowland mixed dipterocarp forest (0°45′N, 110°06′E) in West Kalimantan (Kohyama et al., 2003). We used censuses from 1992 and 1995 (3 years). We also analysed data from Ulu Gadut old-growth foothill dipterocarp forest in West Sumatra (Pinang Pinang plot of 1 ha at 0°55′S, 100°30′E). Here we used censuses from 1984 to 2004 (Kohyama et al., 2019). We also analysed data from old growth warm-temperate rain forest using 1998 and 2008 plot census data from the Segire and Koyohji basin (30°20′N, 130°50′E; 0.89 ha in total area) in Yakushima Island, southern Japan (Kohyama, 1993). Data for intact cool-temperate mixed deciduous forest (42°37′N, 141°36′E) was drawn from the 4-ha plot located in Tomakomai, Hokkaido (Kohyama et al., 2019). We used censuses in 1996 and 2006 (10 years). In all plots, we selected species with ≥6 trees surviving over the period with ≥5 cm dbh (but ≥8 cm for Ulu Gadut). We used plot-specific tree height versus dbh allometry, and biomass equations of Niiyama et al. (2010) for the two Indonesian rain forests, Ishihara et al. (2015) for warm-temperate rain forests on Yakushima, and the site-specific ones for Tomakomai (Kohyama et al., 2019).

| Statistical analysis
We fitted species-level relative productivity p (and other turnover rates, l, r and m) to population biomass B (and W max and N), using an NPP = P plot + F. Observed species turnover varied 10-fold or more among the 390 species. Relative production p ranged from 0.0098 to 0.0638/

| Interspecific variation in turnover
year (95% range) and the median was 0.0262/year. Turnover in species biomass (p, l) and stem density (r, m) were all positively correlated; correlation coefficients (R's) were ranging from 0.300 to 0.570 (Table S1). Thus, species with high-biomass turnover showed high stem turnover. In contrast, in the identity-free data (where we recorded 395 'species' ≥100 survived trees), most of these correlations were markedly lower. Their correlation coefficients ranged from 0.069 to 0.375 (Table S1).
The relative production p and species biomass B appear inversely related among the 390 species. The power function model, p = aB b , was selected by fitting Equation 10, that is, c = 0 (Figures 3a and 4a).
The power exponent b for p on B in observed community was −0.128 (with 95% confidence interval, henceforth 'CI', of −0.150 and −0.105, Figure 4a). The aggregated population of the rarer 418 species (accounting for 5.3% of plot total biomass) showed a relative productivity of p = 0.0232/year, which is located above the regression line for the other, more abundant, 390 species (Figure 3a).
In the identity-free community, p-B exponent was −0.218 with CI [−0.239, −0.197], which was significantly more negative than that in the observed community (Figures 3b and 4a). For the observed data, relative biomass loss l decreased with B (Figure 3c), and the exponent −0.106 was not significantly different from the p-B exponent at −0.128 ( Figure 4a). Therefore, the net rate of biomass change, p−l (year -1 ), indicated by coloured dots in Figure 3, showed no clear dependence on species biomass (Figure 3a,c). The sum of rare species showed a higher biomass loss rate than did the highest biomass F I G U R E 3 Interspecific relationship of per species biomass turnover to biomass in the Pasoh data. (a) Relative production p among observed 390 species; (b) relative production among 395 identity-free species; (c) relative loss l among observed species; and (d) relative loss among identity-free species. Dot colours correspond to decreasing (blue) to increasing (orange) biomass in net change rate, p − l. Regression using a power function with 95% confidence interval is shown, and estimated exponent, or log-log slope, is stated in each panel.  Per-capita recruitment r and mortality m also decreased with species biomass B in observed data (Figure 4a). The power exponent of r on F I G U R E 4 Estimated power exponent and 95% confidence interval for interspecific relationships of turnover on species structure metrics in the Pasoh data. Turnover are relative (= per-unit biomass) productivity p and loss l, per-capita recruitment r and mortality m (all in year -1 ); species structure metrics are (a) biomass B (Mg/ha), (b) 99% tree mass W max (Mg), (c) abundance N (ha -1 ). Black and grey for observed and identity-free data respectively B was not significantly different from that of p on B, but the m-B exponent of −0.059 was larger than the p-B exponent of −0.128 (Figure 4a).
Because species biomass B was strongly correlated with maximum tree mass W max , turnover (p, l, r, m) decreased with W max (Figure 4b), and was similar to turnover dependence on B (Figure 4a).
In contrast to the clear link between turnover with B and W max , we found no significant correlation between p and tree density N, and increasing l, r and m with N in the observed data (Figure 4c).
Among the 390 species, 21 species are considered 'pioneers' (Davies et al., 2003). These species generally had larger p for small biomass species, and a steeper decline of p on B; the p-B exponent for pioneers was −0.34 with CI [−0.54, −0.16] ( Figure S1a). We observed no clear variation in the p-B relationship due to interspecific differences in stem wood density when we consider the 210 tree species that appear in the Global Wood Density Database (Zanne et al., 2009; Figure S1b).

| Linkage with population-level leaf mass ratio
Among species, population-level leaf mass ratio, LMR = B L /B, exhibited a negative correlation with above-ground biomass B in both the observed and identity-free community, which was approximated using Equation 12 (Figure 5a,b). Per-leaf-mass productivity, p/LMR = P/B L (year -1 ) increased with species biomass in observed and identity-free forests (Figure 5c,d) though the slope was significantly less for the latter.

| Linkage with individual tree growth and mortality
Tree relative growth rate, g(W), decreased with tree mass W (Figure 6a).
Species with large standardized maximum mass W max tended to grow faster than species with small W max at the same reference sizes up to 36 cm dbh (at 5% significance level). By contrast, relative growth at maximum size, g(W max ), was significantly negatively correlated with W max among species (Figure 6a), because g(W) sharply decreased around W max for most species. Individual tree mortality μ(W) was typically lower at intermediate tree sizes for most species (Figure 6b). Treesize-dependent mortality μ(W) was negatively correlated with species W max over reference tree sizes up to 64 cm dbh, and so was μ(W max ) with W max (Figure 6b, significant at 5%). Population-level relative productivity p was closely positively correlated to individual tree relative growth at W max , g(W max ) (Figure 6c, R 2 = 0.813 in log-log linear regression). Similarly, the population-level relative loss l was positively correlated with tree mortality at W max , μ(W max ) (Figure 6d, R 2 = 0.592).

| Productivity dependence on subplot-byspecies structure
To evaluate spatial variation in the interspecific productivity-biomass relationship, we calculated the biomass and relative production of local species populations in two hundred 0.25-ha subplots. Selected   11 without B subplot dependence, that is, parameter d = 0 (Figure 7b).

| Plot-level net primary productivity
Estimated species-sum absolute productivity for the Pasoh plot P plot was 10.0 Mg ha −1 year −1 , with a period-mean plot biomass B plot of 468 Mg/ha. The rate of fine litter fall F was 6.49 ± 1.35 Mg ha −1 year -1 (of which leaf litter was 69%). The estimated plot-scale net primary production (NPP) was 16.5 Mg ha −1 year -1 , of which fine litter comprised 39%. Table S2 summarizes the different values estimated using conventional approaches that ignore interspecific variation.

| Interspecific productivity variation in other forests
Analysis of repeated census data from four other old-growth forests showed that relative productivity p declined with species biomass B ( Figure S2). Despite the difference in plot biomass and productivity among forests, the distribution of species on the (p, B) coordinates did not show clear separation between those forests.

| D ISCUSS I ON
We examined how constituent tree species populations with varied structural properties and demographic characteristics contribute to stand-level properties of net primary productivity in species-rich forests. We found that for the 390 most abundant species in the Pasoh 50-ha plot, species with larger population-level biomass also tended to be those with larger maximum individual tree biomass (cf.
Hypothesis 1). Species-level relative production p was lower when species above-ground biomass B was higher and vice versa (cf. Hypothesis 2, Figure 3a). We also found that the sum of rarer species (<100 stems per plot) showed higher productivity compared to highbiomass species (Figure 3a). The proportion of leaf mass in aboveground biomass was lower among populations with higher species overall biomass (cf. Hypothesis 3, Figure 5a). A negative p-B relationship reflects that larger stemmed species typically possess lower relative growth at their largest stem sizes than smaller stemmed species (cf. Hypothesis 4; Figure 6a). Comparable relationships were observed in four other old-growth forests suggesting similar processes.
The variation in per-species turnover in biomass (relative productivity and loss) and abundance (per-capita recruitment and mortality) was positively correlated in the Pasoh Forest. Species turnover in biomass and abundance were negatively related to species biomass B and maximum tree mass W max (Figure 4a,b), suggesting that   Nascimento et al. (2005) noted that abundance turnover rates (recruitment and mortality) were negatively correlated with mean tree size among 95 Amazonian tree species.
In some respects, the patterns we found among co-occurring tree populations in species-rich forests resemble those seen among dominant life forms across various biomes, where populations of organisms that reach larger dimensions have slower turnover than those with smaller body size (Brown, Gillooly, Allen, Savage, & West, 2004;Niklas & Enquist, 2001;Price, Gillooly, Allen, Weitz, & Niklas, 2010). Within-community trade-off between turnover and adult stature would partly reflect general life-history constraints shown by these studies. In addition, within-community interspecific productivity partitioning is also regulated by interactions among individual-stems in the shared habitat. For example, we found small-stature species with low B show lower productivity per unit leaf mass than large-stature species (Figure 5c). This is presumably because small species typically have less access to light on average.
Our examination of the variation in production among subplot-scale species populations suggests that species productivity p in a local subplot is negatively correlated to both local species-level biomass and species-sum subplot-level biomass (Figure 7a), but not in identity-free forest where variation in demography is (we presume) independent of local stand conditions (Figure 7b). Higher p at the same species biomass in less-crowded local stands indicates that the productivity of species populations is regulated by local-scale available resources, such as light.
Our study clarifies why large species tend to show low relative productivity regardless of the well-established fact that trees of larger stature species tend to grow faster than those of smaller stature species over a range of reference stem diameters (Iida et al., 2014;King, Davies, et al., 2006;King, Wright, et al., 2006;Kohyama et al., 2003Kohyama et al., , 2015Lieberman et al., 1985;Poorter et al., 2008;Rüger et al., 2018). The higher growth observed for larger stature species likely reflects that their juveniles tend to appear in less shaded conditions than stems of smaller stature species (Sheil, Salim, Chave, Vanclay, & Hawthorne, 2006), and that shorter species favour higher reproductive allocation rather than vegetative growth at the same stem sizes (King, Wright, et al., 2006;Kohyama et al., 2003). By comparing species at W max , we obtained contrasting patterns of interspecific variation. Because of the generally marked reduction of relative growth observed at large stem sizes, larger sized species also experienced lower g(W max ) than smaller sized species (Figure 6a). In contrast, stem mortality of larger sized species was low at the same stem sizes, and their μ(W max ) compared to smaller sized species was generally lower, even though mortality tended to increase when species approached their largest stem sizes (Figure 6b). Fast growth and generally low mortality (at most sizes) for juveniles of large-stature species result in a low mortality-to-growth ratio, which allows a species population to reach larger maximum sizes (Kohyama et al., 2015). However, close to their maximum size, larger sized species show slower relative growth and survive longer than smaller sized species, resulting in similar negative power exponents between p-B and l-B relationship in the observed (but not in the identity-free) community (Figures 3 and 6).
The variation in demographic characteristics among coexisting tree species has been related to horizontal and vertical aspects of forest structure. Variation in understorey light enhances interspecific differentiation in juvenile growth, survival and shadetolerance sometimes called 'the growth-survival trade-off' (Rees, Condit, Crawley, Pacala, & Tilman, 2001;Wright et al., 2010;Zhu et al., 2018). In contrast, in terms of the vertical structure of the forest canopy, we observe interspecific differentiation between short stature species exhibiting high reproductive allocation for frequent recruitment and tall stature species showing high allocation to vegetative growth for height gain (and delayed reproduction), which is called 'the recruitment versus stature trade-off' (Falster et al., 2017;Kohyama, 1993;Kohyama & Takada, 2009;Rüger et al., 2018). Our results indicate that this trade-off between turnover and adult stature underlies the relationship between relative productivity and standing biomass. To generalize, we see that large-stature, highbiomass species show slower turnover-that is lower recruitment, relative growth and mortality-than small-stature species. We note that the greater range of species sizes that can occur in taller forests, and the associated diversity of size-dependent species behaviours this permits, explains why taller forests tend to be both more productive, and richer in species, than otherwise comparable forests of smaller stature (Duivenvoorden, 1996;Huston, 1994;Sheil & Bongers, 2020).
Improved methods that avoid major biases offer new insights that were not previously available. For example, we can now recognize the contribution that small-stature tree species make to both diversity and stand dynamics. High tree species diversity in tropical forests is known to be largely related to the diversity and persistence of small-sized species (King, Wright, et al., 2006;Niklas, Midgley, & Rand, 2003). We observed that the abundance of small-sized species with high relative productivity collectively contributes to high tropical forest net primary productivity-a contribution which was previously unnoticed due to the biases in previous 'conventional' production estimates (Kohyama et al., 2019).
Viewed in their totality, our results show how demographic differences among species contribute to general properties of ecosystem functioning and biodiversity maintenance. Small-stature, low-biomass species contribute to high forest net primary productivity and enhance ecosystem resilience through rapid biomass turnover and replacement via high per-capita recruitment and relative productivity. In contrast, large-stature species contribute to the persistence of large biomass storage by their longevity (low mortality). The mass ratio hypothesis posits that species-level biomass in a plant community is proportional to species' absolute (not relative) rates of primary production and nutrient supply to soil heterotroph, and thus high-biomass species disproportionately control ecosystem functioning (Grime, 1998). Studies that consider the role of functional diversity have suggested that ecosystem properties predominantly reflect the characteristics of higher biomass species (Finegan et al., 2015;Fotis et al., 2018;Prado-Junior et al., 2016). In contrast, our study implies that the contribution of biodiversity to ecosystem functioning is more complex and multidimensional with different populations making distinct and disproportionate contributions to different ecosystem properties. For example, ecosystem resilience and biomass recovery depend disproportionately on species with rapid turnover while standing biomass and total carbon sequestration depend disproportionately on species with slow turnover.

ACK N OWLED G EM ENTS
We acknowledge that the project of Pasoh 50-ha Forest Dynamics revised the conceptual framing and all authors approved the paper for publication.

PE E R R E V I E W
The peer review history for this article is available at https://publo ns.

DATA AVA I L A B I L I T Y S TAT E M E N T
Observed and identity-free dataset of the Pasoh plot, and source code for estimation of population structural variables and turnover variables are provided in the GitHub repository and has been archived on Zenodo https://github.com/kohya mat/p-B; https://doi. org/10.5281/zenodo.3966750 .