Considering random effects and sampling strategies improves individual compatible biomass models for mixed plantations of Larix olgensis and Fraxinus mandshurica in northeastern China

Abstract

Mixed-species plantations, which are more resilient and productive than pure plantations, have received increasing attention and have gradually become a focus of forest research. Therefore, it has become imperative to develop models to quantitatively describe specific stand types. In this study, biomass models were developed using 105 destructively sampled Changbai larch (Larix olgensis) and Manchurian ash (Fraxinus mandshurica) trees from several mixed-species plantations in northeastern China. In the case that the influence of stand variables and climate factors on tree-level biomass was not significant, seemingly unrelated regression (SUR) models that included tree height (H), crown length (CL) and plot-level random effects were developed to predict tree-level components and total biomass. The results showed that the introduction of plot-level random effects to create seemingly unrelated mixed-effects (SURM) models improved the model fitting, with the determination coefficient increasing by 0.76%∼17.35% (Changbai larch) and 0.60%∼8.15% (Manchurian ash) for all components. Among the developed models, the plot-level SURM model achieved better mean absolute percentage error (MAPE) values of all components (7.50%∼22.12% for Changbai larch and 7.52%∼39.94% for Manchurian ash) according to leave-one-out cross-validation (LOOCV), with significantly lower values than those of SUR models. Two calibration strategies and various numbers of trees (0, 2, 4, 6 and 7) were used to calibrate the random effects model. Model performance improved as the number of trees used for calibration increased, but gains in performance then decreased gradually. When wood and bark biomass were used for calibration, the prediction accuracy of other components (branches, foliage and roots) tended to be uncertain. Overall, it is recommended to use an SURM model with diameter at breast height (DBH) based on more than four local trees and an SURM with DBH, H and CL based on two local trees for random effect calibration for measurements of all biomass components in mixed-species plantations of Changbai larch and Manchurian ash.