Accurate estimations of a tree stem’s heartwood diameter (HD), sapwood width (SW), bark thickness (BT) and diameter outside bark (DOB) are essential for volume and biomass calculations, forest management, and optimal timber production. Understanding these components also provides valuable insights into the ecological health and functioning of forest ecosystems. The additive model system, enhanced by machine learning algorithms, offers new tools for estimation. We constructed parametric taper models and nonparametric multilayer perceptron models to compare the accuracy of HD, SW, BT, and DOB estimates based on 3199 upper stem diameter measurements obtained from 103 sample Pinus koraiensis Siebold & Zucc. trees in Northeast China. In the context of the additive model system for tree stem internal components, we utilized nonlinear seemingly unrelated regression (NSUR) as a parametric method and multitask learning (MTL) as a nonparametric method. In the comparison of additive model systems, MTL (nonparametric) performed better than did NSUR (parametric) by improving the accuracy and generalizability of the stem and internal component estimates. The coefficient of determination (R2) for DOB, HD, SW, and BT increased by 0.31%, 1.39%, 4.21%, and 4.31%, respectively. Based on these results, we conclude that the MTL provides a valid option for improving HD, SW, BT, and DOB and enhancing the volume prediction of stem components.
Read full abstract