2.4.1 Identify the main influencing factors for AGB temporal
stability and productivity in planted and natural forests
We used generalized least squares models (GLS) and LME to assess the
joint effects of climatic, forest structure, and environmental factors
on stability and productivity. Based on the AIC (Akaike’s information
criterion) for GLS and LME models (Zuur et al., 2009), we fit the joint
effects using GLS models with a Gaussian error structure with the
“nlme ” (v3.1-150) software package (Pinheiro, 2020). All
environmental and forest structure variables and climatic variables were
z-transformed to facilitate interpretation of parameter estimates. We
also tested for multi-collinearity; predictors with sufficiently low
variance inflation factors (VIF<5) were included in the model.
We carried out a model-averaging procedure on the basis of AIC
(ΔAIC<4) to decide parameter standard coefficients (Grueber et
al., 2011) for the main influencing factors of stability and
productivity for planted and natural forests using a dredge function in
the “MuMIn” package (Barton, 2015). We considered spatial
autocorrelation (SCV) based on latitude and longitude in the GLS model
(Rousset et al. 2018). All response and prediction variables were
calculated as averages in the continuous investigation cohort. The
structure of our GLS model is as follows:
Stability/Productivity~θ clim+θ fore+θ envi+SCV
+e ,
where θ clim are the climatic variables (annual
precipitation, MAT), θ fore are the mean values of
variables describing forest structure (canopy cover, stand age,
abundance, richness, DBH, tree height, and tree density), andθ envi are environmental context variables
(latitude, altitude, aspect, slope, and soil depth); SCV refers to the
spatial correlation variance structure. e represents the
residual. All terms were modeled as additive effects, and no
interactions were calculated in this top model.
Considering potential nonlinear responses of AGB stability and
productivity to different factors, we used the “mgcv” R-package
to fit the effect of forest structure in generalized additive mixed
models (GAMM) (Fig. S3) (Wood, 2017). The effects of different factors
in the GAMM and GLS were then compared.