During plant growth and development, biomass is accumulated in the form of organic matter through plant photosynthesis, where additional biomass is distributed into the stems to increase their length to obtain more light, toward potentially gaining a competitive advantage over other trees (Nam et al. 2018, Bayen et al. 2020). In this study, the biomass of the stems of six shrub and small tree species exceeded that of the branches and leaves, and accounted for more than 50% of the total aboveground biomass, except for Acer ginnala, which was consistent with many previous results (Singh et al. 2011, Bayen et al. 2020).
Based on the established six specific species and one mixed species model, D, D2H, and ρDH were the best factors for forecasting the aboveground biomass of shrubs. As distinct from Vu Thanh Nam’s method (2018), we employed a single independent variable (single variable D, H, Ca, ρ; compound variables D2H, ρDH) as a predictor in the model, which made the process simpler and more convenient in operation. We observed from the optimal model, even though we employed a single variable, that it still had a satisfactory prediction accuracy, and the R2 of all models was > 0.6 (Table 3).
In terms of model fitting accuracy, Ca did not perform well (Table. S1), and was shown to be inconsistent with some research (Conti et al. 2013, She et al. 2015, Yang et al. 2017). This was because, unlike desert shrubs and subtropical grassland plants (She et al. 2015, Bayen et al. 2020), the crown structures of subtropical shrubs and small trees are irregular in their natural state, which translates to a decrease in the capacity of Ca to predict the biomass of shrub branches and leaves (Poorter et al. 2012, Liu et al. 2015). However, the addition of H to the models developed for the shrubs was reasonable; it improved model accuracy when combined with D and ρ, although the correlation between the H and biomass was lower than Ca (Table 2, Table 3 and Fig. S1). Consistent with previous studies, D2H was one of the best predictors of shrub and small tree biomass (Alvarez et al. 2012, Liu et al. 2015, Dou et al. 2019).
When another variable (ρDH) was employed to predict biomass, the model was better than D and H, particularly for shrub mixed multi-species models. According to previous studies, ρ was one of the most important characteristics of tree species, which had obvious differences between various species. We introduced ρ as a predictor directly into the model with reference to previous research methods; however, the results were not satisfactory (Table. S1). Compared with D and H, the intraspecies variation in ρ was negligible and could be regarded as almost constant (Francis et al. 2017, Nelson et al. 2020). Thus, we attempted to create a new combined entity (ρDH) as a predictor variable.
From the results, the introduction of ρ reduced SEE and increased R2, which indicated that the inclusion of ρ enhanced the accuracy of the model, which aligned well with the work of Ali (2015). Many studies have indicated that ρ is an essential factor for improving model accuracy (Yepes et al. 2016, Kebede et al. 2018). Furthermore, it was indicated that ρ had enhanced relevance in mixed models, as it was believed to augment the differences in the physiological structures and functional characteristics of tree species (Xu et al. 2015, Nam et al. 2018). Taking ρ as a portion of the independent variable can reduce the errors caused by D and H in the measurement to a certain extent, as well as the influences of morphological differences on the accuracy of multi-species models (Pilli et al. 2006, Zeng et al. 2017).
Comparing the accuracy of a model in estimating the biomass of a component of the same tree species, the allometric model possessed a higher predictive ability for wood organs (stems and branches) than leaves. Some studies attributed the low predictive power of leaf biomass models to the ephemeral nature of leaves and their destruction by herbivores (Roxburgh et al. 2015, Sanquetta et al. 2015, Bayen et al. 2020). The accuracy of the allometric leaf model of evergreen shrubs in this study was higher than that of deciduous shrubs, which appeared to confirm this assumption.
Due to the difficulties involved with fully excavating the root biomass (particularly fine roots), only the aboveground biomass was considered in our research, which might have translated to limitations in the application of the equations. We acknowledge that our developed models were based on a small number of individual samples (i.e., 32 to 38 individuals per species). Therefore, we do not recommend the use of predictive models for tree species beyond the range of predictor variables, which will likely cause significant errors in the estimated values.