Abstract: [Objective] The traditional method based on forest inventory data plays an import role in assessment of forest biomass at regional scale and its dynamics and verification of the remote-sensing based model and improvement of its prediction precision. The forest biomass methods at regional scale have attracted most attention of researchers, developing the stand-level biomass model has become a new trend. Based on 1990-2010 forestry inventory data (the 5^th inventory) of natural larch forest in the east of Daxing'an Mountains, we studied how to use two methods (i.e., stand biomass models respectively based on stand variables and stand volume) to establish the additive system of stand-level biomass equations, and analyzed their prediction precisions. These provided technical and theoretical support for accounting and monitoring the Chinese forest biomass and carbon stock.[Method] Structure of model errors (additive vs. multiplicative) of total and component biomass allometric equations in two stand-level biomass of natural larch forest in the east of Daxing'an Mountains were evaluated using the likelihood analysis, and aggregation system was used to establish the stand-level biomass additive models, while nonlinear seemly unrelated regression was used to estimate the parameters in the additive system of biomass equations. The stand-level biomass model validation was accomplished by Jackknifing technique in this study.[Result] The assumption of multiplicative error structure was strongly supported for total and tree components biomass equations in two stand-level biomass additive systems. Thus, the additive system of log-transformed biomass equations should be developed. The adjusted coefficient of determination (R_a²) of two stand-level biomass additive systems for natural larch forest in the east of Daxing'an Mountains were above 0.94, the mean relative error (ME) was between 0%-5%, and the mean absolute relative error (MAE) was less than 15%. All the precisions of total and tree components biomass equations in stand-level biomass additive system were above 98%.[Conclusion] Although the significance of likelihood analysis was used in individual tree biomass equations by several studies, it has not been widely applied in stand-level biomass equations. In addition, in order to estimate model parameters more effectively, the additivity of total and components biomass should be taken into account. Overall, there were differences between the two methods, but good precisions of the two methods were obtained. The two methods would be suitable for predicting the stand-level biomass of natural larch forest in the east of Daxing'an Mountains.

Key words: natural larch forest, stand-level biomass, error structure, likelihood analyses, additive system