关键词: “迎春5号”杨树, 似乎不相关回归, 混合效应模型, 生物量, 模型校正

Abstract: Populus nigra × P. simonii is one of the important tree species with the characteristics of fast-growing and high-yield. The accurate estimation of its biomass is of great significance. Based on diameter at breast height (DBH) and tree height (H), we constructed two seemingly unrelated regression models (SUR) and seemingly unrelated mixedeffects models (SURM) introduced random effects of plots, respectively (SUR1 and SURM1: DBH as the independent variable; SUR2 and SURM2: DBH and H as independent variables), and analyzed the prediction accuracy of SURM models. The results showed that after adding H to the model, the simulation effect on the biomass of trunk, branch, and leaf was significantly improved (R² in the SUR model increased by 0.1%-1.6%, and R² in SURM increased by 0.6%-7.0%), but the improvement on bark and root was poor. After considering the random effects at the plot level, the fitting effect of branch, leaf, and root biomass in the model was significantly improved, with the R² of each component in the SURM model increased by 5.7%-16.5%, while the improvement effect of trunk and bark biomass simulation was small. In addition, the mixedeffects model using the plot level (SURM1) performed better than using two variables (SUR2) to predict biomass, and the fitting effect of each component of the SURM1 model was better than the SUR2 model (R₂ increased by 0.2%-9.9%). The SURM model was more accurate for the biomass fitting and prediction of each component than the SUR model. In practice, when using the SURM model for biomass prediction, considering the difficulty of obtaining biomass data, it was determined that four trees were randomly selected for the correction of random effects. The two types of SURM models proposed here were applicable to different data types, which can provide basic support for accurately assessing the biomass of Populus nigra × P. simonii.

Key words: Populus nigra × P. simonii, seemingly unrelated regression, mixed-effects model, biomass, model correction