Multiple Drivers of Biomass Change in Subtropical Natural Forests


 Forest ecosystems play an important role in regulating the global carbon, a substantial portion of terrestrial carbon pool which is stored in biomass stocks. However, how multiple biotic (i.e. topography) and abiotic (biodiversity, stand structure, and functional traits) influence forest biomass in natural forests, the relative important of these factors determine biomass is still controversial for subtropical natural forests. We used forest inventory data from nine 1-ha plots at different altitude gradients in China’s subtropical forests. We used multiple analyse to quantify the relative importance of multiple facets of diversity, key functional traits, stand structural attributes, and topography variables in determining forest biomass. We found that multiple facets of diversity and stand structure variables enhances biomass. Specifically, large-diameter trees had a strong positive effect on biomass and were the most important factor in determining biomass. Plant functional traits were closely related to biomass. Community-weighted mean value (CWM) of maximum height positively correlated with biomass, but CWM of wood density negatively correlated biomass. Topographic factors including elevation and slope, had a positive effect on biomass. Moreover, among the aforementioned four types of variables, stand structure had the greatest impact on biomass and is linked to diversity-biomass relationship. Topography mainly indirectly affected biomass by altering multiple diversity and stand structure. Functional traits also directly and indirectly affected biomass. Overall, these results support niche complementarity effect and mass-ratio hypothesis. Our results indicate that biodiversity is essential for maintaining ecosystem functions of species-rich subtropical natural forests. Further, adjusting stand structure may be an effective forest management approach to increase forest carbon storage.


Introduction
Forest ecosystems play an essential role in biodiversity conservation and carbon cycling (Pan et al. 2011;King et al. 2012). Species-rich subtropical forests cover an extensive area and are crucial in regulating global carbon cycling and maintaining high biodiversity . Evidence is mounting that diverse forest communities generally accumulate biomass more rapidly than species-poor ones (Jucker et al. An alternative but not mutually exclusive mechanism to the niche complementarity is the mass-ratio hypothesis (Grime 1998), which states that the most dominant species in the community drive the ecosystem processes by means of their traits. Moreover, species with acquisitive traits lead to faster carbon capture ability, while species with conservative traits possess a higher long-term carbon sequestration strategy (Díaz et al. 2009). Plant functional traits are the key aspects shaping forest biomass dynamics (Díaz et al. 2007; Lohbeck et al. 2015). Species-level differences are important in structuring highly diverse communities (Kraft et al. 2008). Functional trait trade-offs are useful metrics for understanding community response to global change (Kimball et al. 2016). In addition, previous studies reported that plant biomass accumulation predicted by phylogenetic diversity is stronger than by species richness and functional diversity (Cadotte et al. 2008;Liang et al. 2019).
In addition to species diversity, stand structural attributes in natural forests also have a strong in uence on biomass and may interfere with the relationship between species diversity and biomass. Structural variability may in uence ecosystem processes and functioning. Stand structural complexity increases light capture ability, light-use e ciency, plant water and nutrients use e ciency, promoting the accumulation of biomass in forest ecosystems (Hardiman et al. 2011). Forest biomass is intrinsically related to tree size. A study in natural boreal forests found that tree size inequality links to diversity and aboveground biomass and regulates above-ground biomass and species diversity by interactions among individuals (Zhang and Chen 2015). Studies at global and regional scales have shown that largediameter trees comprise a large fraction of the stand basal area and biomass in many forests (Paoli et  It is noteworthy that environmental variation is also a key regulator of productivity/biomass in forests. Topography, for example, represents many aspects of microenvironmental changes. Topographic characteristics such as elevation, slope, and aspect in uence microclimate, which are known to drive tree ). In addition, soil type, soil water potential, and nutrient cycling are affected by topography, affecting tree biomass accumulation. Microclimate created by topography also changes stand attributes and leaf characteristics.
In this study, we set up plots at different elevation gradients in three subtropical forests. This ultimate goal is to determine how multiple diversity, plant functional traits, stand structure attributes, and topographic factors affect biomass in subtropical forests. Speci cally, we address four questions. First, we ask (i) is there a positive relationship between multiple diversity and biomass in a subtropical natural forest? Second, we incorporate data on key plant functional traits related to tree growth to ask (ii) is biomass in uenced through the mass-ratio effects? Third, we ask (iii) how do stand structural attributes affect biomass of subtropical forests and maintain the diversity-biomass relationship? Finally, we ask (iv) how do topographic factors drive biomass other biotic factor? To answer these research questions, we used bivariate relationships, multiple linear regression, variation partitioning analysis and structural equation model based on existing research theories (Fig. 1a) to quantify the relative importance of multiple diversity, functional traits, stand structure, and topographic factors as the best predictors of variation in biomass.

Plot settings and data collection
The forests in these three Nature Reserve have been well protected and with virtually no human disturbances or re for a long time. The canopy is completely closed. From 2018 to 2020, we established three 1-ha study plots in different altitude gradients of the three Nature Reserves, a total of nine 1-ha forest plots were set up for vegetation census. Each forest plot consisted of twenty-ve 20×20 m subplots. During the inventory, we recorded the geographic coordinates of each plot and the elevation of each subplot. Within each subplot, we followed the same procedures to map, tag, measure, and identify all woody-plant individuals with a DBH ≥ 1 cm. We also recorded the species names, DBH, height, crown width, health status, and spatial coordinates. The spatial coordinates of each individual had twodimensional accuracy of ± 15 cm. A total of 33,172 individuals were recorded in the nine 1-ha plots, belonging to 343 species, 142 genera, and 61 families.

Variables used in analyses
We estimated standing biomass for live trees each species (tons, t) in the 20×20m subplots using allometric equations based on DBH. The total biomass for each tree included leaves, branches, stems, and roots. We then calculated standing biomass, which represents the accumulated productivity since the stand establishment. Species richness or the number of species in a given area, a measure of changes in species dynamics, is the most common measure for characterizing community diversity. The oftenrecommended Simpson and Shannon-Wiener indices may stall be highly correlated with species richness. Due to the strong correlation between log species richness and these indicators (Shannon: r = 0.83, Simpson: r = 0.55, with n = 225 of subplot), species richness was selected as a key representative. We also calculated tree Faith's phylogenetic diversity based on specie's evolutionary distances (Faith, 1992), which incorporates relative abundances and phylogenetic distances at set spatial scales. Further, trees with present branch lengths were generated according to Zanne et al. (2014), based on APG III. The generated phylogenetic tree is shown in Figure S1.
Functional diversity was quanti ed using the functional dispersion index (Laliberté and Legendre 2010).
We calculated the functional diversity for each subplot using the wood density and maximum tree height for each species. Wood density data was extracted from the global wood density database ; https://datadryad. org//handle/10255/dryad.235). Tree maximum height for each species was obtained from the Flora of China (http://iplant.cn). Figure S1 shows wood density and maximum height for each species used in this study. We calculated the community-weighted means for wood density . We calculated the the coe cient of variation (CV) to represent tree size variation, which is the ratio of standard deviation of all DBH measurements to the mean DBH within each subplot. Stand density and maximum diameter were used to represent the number of individuals and dominant species in each subplot. Topographic variables included elevation, slope, and convexity.
According to the elevation of each subplot, the slope and convexity within the subplot were calculated. Convexity was calculated as the elevation of the central subplot minus the average elevation of the eight adjacent subplots.

Statistical analyses
With the ln-transformed biomass, we rst tested correlations between individual continuous predictor variables and forest biomass using Pearson correlation coe cients. All indicators were standardized before the analysis. We used multiple linear regression to evaluate the effects of all predictors simultaneously. The full model included multiple diversity (species richness, phylogenetic diversity, and functional diversity), functional traits (CWM WD , CWM MH , and Rao's Q), structural variables (stand density, maximum DBH, and DBH variation), and topographic variables (elevation, slope, and convexity). The variance in ation factor (VIF) was used to diagnose the multicollinearity of the explanatory variables, with VIF > 10 indicating excessive collinearity. The Akaike information criterion (AIC) was used to compare the model results. The VIF was calculated with the R package 'CAR' (Fox and Monette 1992).
The comparison and averaging of models were conducted using the R package 'MuMIn' (Bartoń 2016).
We conducted a random forest classi cation analysis to identify the main predictors of forest biomass. The random forest analysis allowed us to identify the most important drivers of biomass among 12 variables. This analysis accounted for interactions and nonlinear relationships between predictors, and addressed the multicollinearity problem in multivariate regression. The t for each tree was determined by randomly selecting cases. In addition, the importance of each predictor variable was estimated from the percentage increase in the mean square error between observation and prediction, and the decrease was averaged over all the trees to produce the nal estimation of importance. This accurate measure was computed for each tree and averaged over the forest (i.e. 9999 trees) using the R package 'randomForest'.
Variation partitioning analysis was used to quantify the relative importance of multiple diversity diversity, functional traits, stand structure and topographic variables as predictors of biomass. In particular, the main goal of these analyses provides insights into whether these four types can explain a unique portion of the variance, which further explains the relative importance of the two underlying mechanisms. The variation partitioning analyses were conducted using the R package 'vegan' (Oksanen et al. 2017). All the above analyses were run in R version 3.3.3 (R Core Team 2017) (http://www.R-project.org/).
We rst performed a principal component analysis was used to reduce the number of multiple diversity, functional traits, stand structure and topographic variables. Variables were standardized for the principal component analysis. We used the structural equation model (SEM) to investigate the impact of these four types of predictors on biomass directly and indirectly. We rst designed a full conceptual model framework that included all possible pathways based on existing research theories. Path coe cients were obtained by using maximum likelihood estimation. We used χ 2 test, AIC, and the root mean square error (RMSE) of approximation to evaluate the model's tness. All SEM analyses were performed using AMOS 21.0 (Amos Development Corporation, Chicago, IL, USA).

Relative importance of predictor variables on biomass
The nal multiple regression model explained 67.35% of the variation in biomass (Fig. 3a). Among the all stand structural variables, the maximum DBH had the strongest effect on biomass, followed by stand density and DBH variation. Among the functional traits, CWM MH had a positive effect on biomass, while CWM WD and Rao's Q had a negative effect on biomass. Moreover, functional diversity and elevation had a positive effect on biomass. While species richness, phylogenetic diversity, and slope did not appear in the nal model (Fig. 3a). Meanwhile, random forest model explained 62.79% of the variation in biomass. The most relative importance was the maximum DBH, followed by species richness, elevation, DBH variation and CWM MH . The relative importance of the remaining indicators was relatively low (Fig. 3b).

Direct and indirect effects of main drivers on biomass
Variation partitioning analysis revealed that the single effect of stand structure explained a much greater portion of variance in biomass (29%), followed by the combined effect of stand structure and multiple diversity explained 11%, respectively. Topography, multiple diversity and stand structure jointly explained 7% of biomass. The single effect of functional traits and the combined effect of stand structure and functional traits all explained 5% of biomass, respectively. The combined effect of topography and functional traits also explained 5% of biomass (Fig. 4a).

Discussion
Multiple facets of diversity promotes forest biomass A substantial body of evidence sustains that biodiversity enhances productivity or biomass in forest ecosystems (Tilman et  As expected, we found that species richness, phylogenetic diversity and functional diversity were positively related to biomass (Figs. 2d-f). The SEM showed that multiple diversity also increased biomass directly (Fig. 4b, c). Our results supports the idea that the niche complementarity is operating in species-rich subtropical natural forests. In addition, the relationship between species richness (or phylogenetic diversity) and biomass was not statistically signi cant when other factors were accounted for in the multiple regression model (Fig. 3a), indicating that species richness and phylogenetic diversity have independent effects on biomass, which is consistent with previous studies in temperate forests Many studies generally nd that the values of community-level traits respond to environmental gradients. We selected two key functional traits to re ect tree growth, species composition and response to environmental conditions. Wood density is a key trait driving the trade-off between growth and survival, as low wood density allows rapid growth of the canopy, higher diameter growth rate, and greater productivity, whereas high wood density results in a greater chance of survival ). We found that the negative correlation between CWM WD and biomass suggests that plots with lower CWM WD have higher biomass accumulation rates (Fig. 2h). . Because tree biomass is dependent on tree size, our study indicated a signi cant positive relationship between maximum DBH and biomass (Fig. 2b). To be sure, our results suggest that maximum DBH had the greatest effect on biomass compared to other predictors.
Multiple analyses showed that the DBH variation was positively correlated with biomass (Figs. 2-3), suggesting that tree size inequality was the main regulation mechanism for biomass. Tree size inequality is caused by both differences with and among species and represents niche differentiation and facilitation (Coomes et  The SEM revealed that stand structure directly and strongly increased biomass and became the most important factor explaining variation in biomass. Notably, multiple diversity indirectly increased biomass via increasing stand structure (Fig. 4b,c). Collectively, this study suggests that a structurally complex forest stand increases biomass of subtropical natural forests and plays a major role in modulating diversity-biomass relationships. That is, these results demonstrate the complementarity effect in subtropical natural forests. Previous results have found that the positive diversity effects on forest biomass were mediated by promoting tree size inequality in boreal forests (Zhang and Chen 2015). Our nding is consistent with the reported variations among individuals in natural forests, critical to species coexistence (Clark 2010).

Effect of topographic factors on forest biomass
Topography can in uence multiple diversity, functional traits, stand structural attributes, and biomass in China's subtropical forests in multiple ways. Topography is strongly correlated with soil nutrient availability, temperature, hydrology, and light conditions (Mascaro et al. 2011;Tateno and Takeda, 2003), and thus is expected to play an important role in shaping the relationship between biodiversity and ecosystem functioning (Grossiord et al. 2014; Liu et al. 2014). Here, our results showed that diversity and biomass decreased with increasing elevation gradients. In contrast, our results found that species diversity and biomass increase with altitude gradient (Fig. 2j; Table S1), highlighting the role of habitat heterogeneity in regulating variation of diversity and biomass in subtropical natural forests. An increase in variation in aboveground live tree biomass with increasing elevation gradients has been observed in a central Amazonian (de Castilho et al. 2006).
Increasing temperatures are considered to be considered to be a key driver of upward migration of vascular plant species in mountains (Gottfried et al. 2012). A oristic composition study of alpine summit vegetation recorded that while plant species richness continued to increase, the trend in the upward shift of alpine plants accelerated (Walther et al. 2005). Furthermore, with the introduction of alien species, species richness of the region will usually increase (Sax and Gaines 2003), and due to the current warming of the cold and humid regions, the species richness is also expected to increase (Pauli et al. 2012).Moreover, due to climate change and global warming, some species may move to high altitude areas that were previously not suitable for plant growth, suggesting a widening of the adaptation range of such species. These reasons have also caused forest biomass of this area to increase with altitude.
Additionally, functional traits can be used to evaluate functional responses to restoration projects (Laughlin, 2014). First, we found that elevation (or slope) has a signi cant negative effect on CWM WD (Table S1). We preliminarily predict that tree species with low wood density and higher diameter growth rate (e.g. large-diameter trees, fast-growing tree species) are increasing along elevation gradient, tree species with high wood density and lower growth rate (small tree, slow-growing tree species) are reducing change of survival. Second, our results also fund that elevation has has a signi cant positive effect on maximum DBH and CWM MH (Table S1), suggesting that large diameter trees and higher trees are increasing along elevation gradient. Collectively, in response to global climate changes, these functional traits with altitude gradients will affect tree composition, life strategies and individual differences, emphasizing that functional traits respond to changes in elevation gradients that has important consequences for ecosystem functioning. Meanwhile, future studies should consider the impact of environmental changes caused by topography on ecological functions.

Conclusions
This study shows the relative importance of multiple facets of diversity, key functional traits, stand structural attributes, and topographic variables on biomass variation in subtropical natural forests. We found that multiple facets of diversity were positively correlated with biomass. Plant functional traits also had a signi cant effect on biomass. More importantly, stand structure was a driver of biomass and regulated the diversity-biomass relationship. In addition, elevation and slope had a positive effect on biomass. The SEM showed that topography mainly and indirectly affected biomass via altering multiple diversity and stand structure. In particular, the impact of elevation on biodiversity, functional traits and biomass was worthy of attention. Overall, this study advances our understanding of the mechanisms affecting biomass in complex subtropical natural forests. We also emphasize that many aspects of biodiversity are crucial for maintaining biomass of species-rich subtropical natural forests, supporting niche complementarity effect and mass-ratio hypothesis. Stand structure attributes should also be highly valued in the process of forest management, as well as the in uence of environmental conditions on maintaining forest functioning.  Relation between biomass and structural variables (stand density, maximum DBH and DBH variation), diversity indices (species richness, phylogenetic diversity and functional diversity), functional traits (Rao's Q and community-weighted mean (CWM) of maximum height and wood density) and topographic variables (elevation, slope and convexity). Light grey bands represent 95% con dence intervals, and the legend is shown in gure 1b.