Site description and plots design
Our study was conducted in Dawei Mountain area (22°35′-23°07′ N, 103°20′-104°03′ E), located in Yunnan Province, Southwestern China. The annual rainfall is 1700–1900 mm. The annual average temperature is 22.6 ℃. The coldest month (January) is 15.2 ℃ and the hottest month (July) is 27.7 ℃. There is no frost in the whole year. The elevation range between 225 m and 2365 m for the highest peak of Dajian Mountain. The forest types change significantly from the bottom to the top of mountain.
In order to ensure the comparability of forest communities in different elevation gradients, only old-growth forests far away from disturbance were selected. A vegetation and soil comprehensive investigation was conducted before plot selection. With the help of local forestry department, four elevation gradients, containing 800 m, 1200 m, 1600 m and 2000 m, respectively were selected and fourteen 20 m × 20 m plots were set with a distance more than 100 m for each plot in each elevation gradient. Total 56 plots were selected. Between 2017–2018, all the individual trees were identified in the plot or lab through the collected specimen to species level (Li et al. 2019). Details of species composition are listed in Annex 1 of Additional file 1.
Quantification of aboveground C storage
We measured the diameter at breast height (DBH) of all the trees with DBH higher than 5 cm in each plot. The telescopic pole was used to measure tree height lower than 18 m and the clinometer was used for the height measurement for the other trees. The allometric equation, which was according to height, DBH and wood density, was used to calculate the aboveground biomass of individuals whose DBH were higher than 5 cm (Chave et al. 2014). Aboveground C storage was calculated by multiplying aboveground biomass by 0.5 (Chave et al. 2005). Wood density for most species were obtained from field investigation and other species’ wood density were used the wood density of average family or average wood density of corresponding plot (Cavanaugh et al. 2014). The global allometric equation was calculated as follows:
where ρ represents the wood density (g cm − 3), H represents the height (m), DBH represents diameter at breast height (cm).
The quantification of stand structural and species diversity
Species and stand structural (height and DBH) diversity were quantified by Shannon-Wiener biodiversity index (Ali et al. 2016). Recommended by Ali et al (2016), different DBH classes (8, 6, 4, and 2 cm) and height classes (5, 4, 3, and 2 m) were calculated. The proportions of individual species, height class and DBH class was represented by the relative basal area (Finegan et al. 2015; Ali et al. 2016). Because the different discrete height and DBH diversity classes may predict aboveground C storage differently, relationships between aboveground C storage and each class of stand structural diversity were elevated and the lowest AIC values were used to select the classes used for the models in Annex 2 of Additional file 1 (Yuan et al. 2018b).
Functional traits
Leaf nitrogen concentration (LNC), leaf phosphorus concentration (LPC), N:P ratio, specific leaf area (SLA), leaf dry matter content (LDMC) and leaf thickness (LT), which were crucial for plant survival and growth (Wright et al. 2010; Finegan et al. 2015; Ali et al. 2017), were measured of all the species in the plots in both young and mature leaves, on account of the central plant trade-offs strongly correlating with leaf longevity (Aerts et al. 2000; Ali et al. 2017). All the leaf functional traits were measured based on standard measurement methods (Pérez-Harguindeguy et al. 2013).
The calculation of CWM trait values in each plot were based on the following formula (Conti and Díaz., 2013; Ali et al., 2017):
where CWM (traitX) represents the CWM X trait, s represents the species number in each plot, pi represents i th species’ relative abundance in the plot and xi represents the i th species’ trait value.
Functional diversity was calculated as follows (Conti and Díaz. 2013):
Statistical analyses
One-way ANOVA and least square difference (LSD) multiple comparison tests were used to test the difference in aboveground C storage and abiotic factors among different elevations (Annex 3 of Additional file 1). Pearson correlation coefficients were employed to investigate the relationships among aboveground C storage, species diversity, functional dominances, functional diversity, stand structural diversity and elevation in Annex 1 of Additional file 2 (Li et al. 2019). Shapiro-Wilk test was used to examine the normality of all the date (Zhang et al. 2012). Non-normal continuous variables were natural-logarithm-transformed to improve normality and linearity before the date statistics (Zhang et al. 2012; Li et al. 2019).
Simple and multiple linear regressions were employed to explore the relationships between aboveground C storage and explanatory variables (Li et al. 2019). The combination of 28 indices yields total 268435455 models, which were beyond the capacity of R (Ali et al. 2017). To select the optimal subset of predictor variables of the aboveground C storage, ordinary least squares multiple regression analyses was first conducted with all 12 CWM indices, resulting 4095 possible models. Then we used the other 15 biotic indices (FDvar, species diversity and stand structural diversity) to conduct multiple regression analyses, which resulted 32367 possible models. Finally, we combined the elevation with the two former best subsets of predictors to conduct multiple analyses and the 8 predict variables resulted 255 possible models. All possible models were evaluated with corrected Akaike information criterion (AICc) and the best-fit regression model were selected based on the lowest AICc values (Li et al. 2019; Wen et al. 2019). Multicollinearity was diagnosed by variance inflation factor (VIF) and all VIF values were lower than 10 in the best-fit regression model, which suggested our results were not affected by collinearity among predictor variables (Graham 2003). Detailed statistics of all models are provided in Annex 2 of Additional file 2, respectively. The ‘glmulti’ package in R 3.6.1 was used to select models (Calcagno 2013).
Structural equation model (SEM), which was widely used to explore the complex relationships between ecosystem functioning and predictor variables (van der Sande et al. 2017), were conducted to invest how the biotic and abiotic factors affect aboveground C storage directly or indirectly. The variables retained in the best-fit regression model were used to construct SEM (Ali et al. 2017; Wen et al. 2019). The comparative fit index (CFI), goodness-of-fit index (GFI), root mean square error of approximation (RMSEA), Chi-square (χ2) test and AIC were employed to test the fitness of SEM (Zhang and Chen. 2015). The SEM were implemented using the AMOS 21.0 software.