Key points are not available for this paper at this time.
Accurate quantification of stand aboveground biomass (SAGB) is fundamental for assessing forest carbon storage, understanding ecosystem processes, and supporting sustainable forest management, particularly in rapidly growing and widely distributed forest, such as moso bamboo (Phyllostachys edulis) forest system. Biomass estimation of this forest remains a quite challenging because stand structural and environmental factors are often not jointly considered, and complex nonlinear relationships among the influencing factors are difficult to represent. In this study, field survey data from 532 moso bamboo sample plots across eight provinces in southern China were compiled to analyze the regional-scale patterns of SAGB. The important machine learning methods, such as random Forest (RF), support vector machine (SVM), boosted regression trees (BRT), and artificial neural network (ANN) were implemented within a unified modeling framework to develop models using stand structure, soil, climate, and topographic variables as predictors. The results consistently identified basal area (BA) and dominant diameter (DD) as the key determinants of SAGB with strong direct effects, whereas environmental variables, including elevation, soil rock fragments (SRF), and Hargreaves reference evaporation (Eref), significantly influenced SAGB through indirect pathways. Among the evaluated models, RF exhibited the strongest prediction performance with R2 > 0.65, effectively capturing the nonlinear responses and reducing estimation bias across the full range of SAGB values. Analysis of variable importance and partial dependence further confirmed the dominant roles of stand structural attributes. Comparison with the published biomass models demonstrated an improved robustness and regional transferability. This study thus provides a comprehensive framework for improving biomass estimation accuracy, enhancing carbon stock assessment, and supporting the sustainable management of Moso bamboo forests.
Zhou et al. (Sat,) studied this question.