Abstract: Based on data of 36 analytical trees of Pinus sylvestris var. mongolica, we developed univariate and binary additive models for the estimation of biomass and carbon stock. The nonlinear seemingly uncorrelated regression was used to estimate the parameters, while the jackknifing technique was used to evaluate the predictive ability. Covariance analysis was used to eliminate differences between individual trees. Differences among the five approaches estimating carbon stock were tested with ANOVA. All models fitted well. The adjusted coefficient of determination (R_a²) was above 0.90. The mean prediction error (MPE) was between -0.5 and 0.5 kg, the mean absolute error (MAE) was less than 15 kg, and all models had a good fit index (FI>0.88). With the addition of tree height as a variable, the binary model could improve the fitting effect and predictive ability of models for biomass and carbon stock. By comparing different methods, we found that the carbon stock model had obvious advantages in estimating the carbon stock of each organ and the whole tree. The approaches using the carbon concentration constant (i.e. 0.45 or 0.50) produced significant biases in estimating carbon stock of individual trees. In conclusion, the models for the estimation of biomass and carbon stock established in this study can accurately predict the biomass and carbon stock of each organ and individual trees. The error of the binary carbon stock model is smaller when estimating carbon stock of individual trees.

Key words: Pinus sylvestris var. mongolica plantation, carbon stock, aggregate additive model, jackknifing technique.