Abstract:With the development of national-scale forest biomass monitoring work, accurate estimation of forest biomass on a large scale is becoming an important research topic in forestry. In this study, the stem wood, branches, stem bark, needles, roots and total biomass models for larch were developed at the regional level, using a general allometric equation, a dummy variable model, a mixed effects model, and a Bayesian hierarchical model, to select the most effective method for predicting large-scale forest biomass. Results showed total biomass of trees with the same diameter gradually decreased from southern to northern regions in China, except in the Hebei province. We found that the stem wood, branch, stem bark, needle, root, and total biomass model relationships were statistically significant (p-values < 0.01) for the general allometric equation, linear mixed model, dummy variable model, and Bayesian hierarchical model, but the linear mixed, dummy variable, and Bayesian hierarchical models showed better performance than the general allometric equation. An F-test also showed significant differences between the models. The R 2 average values of the linear mixed model, dummy variable model, and Bayesian hierarchical model were higher than those of the general allometric equation by 0.007, 0.018, 0.015, 0.004, 0.09, and 0.117 for the total tree, root, stem wood, stem bark, branch, and needle models respectively. However, there were no significant differences between the linear mixed model, dummy variable model, and Bayesian hierarchical model. When the number of categories was increased, the linear mixed model and Bayesian hierarchical model were more flexible and applicable than the dummy variable model for the construction of regional biomass models.