Construction of Carbon Content Models for Major Coniferous Tree Species in The Northeastern Forest Based on Tree Species Random Effects
To accurately estimate the carbon content of coniferous tree species in the northeastern forest region,in-dividual tree carbon content,which is difficult to measure directly,was inferred using easily measurable forest inven-tory parameters.Based on data from 129 dissected trees of three coniferous tree species in Heilongjiang Province,differences in carbon content among four tree components(i.e.,stem,branch,leaf,and root)were analyzed.Seem-ingly Unrelated Regression(SUR)methods were employed to construct single-variable carbon content SUR models(SUR1)and two-variable carbon content SUR models(SUR2).Additionally,species-level mixed effects were intro-duced into the SUR models to construct Seemingly Unrelated Mixed Effects Regression(SURM)models(SURM1 and SURM2).Random sampling was utilized to select calibration samples for estimating species-level random effects parameters,and evaluate the predictive ability of the model at different sample sizes.The results showed that:(1)The introduction of species-level mixed effects significantly improved the fitting accuracy of models for branches and leaves,with R2a values increasing by over 17%,and by 6%and 13%for stems and roots,respectively;(2)SURM2 outperformed SURM1 in the fitting and prediction accuracy of stem carbon content,with no significant difference observed in the fitting and prediction accuracy of leaf,branch,and root carbon content;(3)When using three randomly selected sample trees to compute species-level random effects,the predictive performance of SURM models was superior to that of SUR models and SURM models considering only fixed effects.Introducing random effects into the basic SUR model system can enhance predictive accuracy.This study recommends using SURM mod-els to predict the carbon content of three coniferous tree species in the northeastern forest region,providing valuable insights for estimating carbon content in this region.
万方数据
维普
