[Constructing biomass models for natural Quercus mongolica based on Bayesian seemingly unrelated regression]

Area of Science:

• Forestry and Ecology
• Quantitative Silviculture
• Biomass Modeling

Background:

• Accurate biomass estimation is crucial for understanding forest carbon dynamics and sustainable forest management.
• Existing biomass models for Quercus mongolica in Heilongjiang Province require refinement for improved precision.
• The influence of tree height (H) on biomass prediction accuracy needs further investigation.

Purpose of the Study:

• To construct and compare biomass models for natural Quercus mongolica using diameter at breast height (D) and tree height (H).
• To evaluate the performance of different statistical methods, including Bayesian estimation, for biomass modeling.
• To determine the optimal predictors and modeling approaches for accurate biomass prediction.

Main Methods:

• Development of biomass models using multivariate likelihood analysis and seemingly unrelated regression.
• Comparison of models based solely on diameter (D) versus those incorporating diameter and height (D-H).
• Application and comparison of various Bayesian estimation techniques (DMC, Gibbs1, Gibbs2, Gibbs3, Gibbs4) against feasible generalized least squares (FGLS).

Main Results:

• Tree height (H) significantly improved stem biomass models (R² increased to 0.988), but not for other components (branch, foliage, root).
• Logarithmic transformation was appropriate due to multiplicative error structures in both D and D-H models.
• Bayesian estimation using local data (Gibbs4) yielded the lowest prediction errors (MAPE as low as 9.8%) and narrowest prediction intervals, outperforming classical methods.

Conclusions:

• Diameter at breast height (D) and tree height (H) are key predictors for Quercus mongolica biomass, especially for stem component.
• Bayesian seemingly unrelated regression with accurate prior information (derived from local data) offers superior estimation stability and uncertainty reduction compared to classical statistics.
• The Gibbs4 approach provides the most accurate and reliable biomass predictions for Quercus mongolica in the study region.