Modelling height to crown base using non-parametric methods for mixed forests in China

Abstract

The height to crown base (HCB) of a tree is a vital characteristic that reflects the self-thinning ability of a tree, and it is used to determine the crown size. and predict the crown recession rate. This study simulated the HCB of Spruce fir broadleaved mixed forest in Northeast China using four non-parametric model approaches: generalized additive model, Cubist, boosted regression tree (BRT), and multiple adaptive regression spline. Because of the different genetic characteristics and growth patterns of different tree species, species-specific tree groups were formed, and the HCB of each species-specific group was simulated by the different models. Relative importance and partial dependence analyses were performed to identify the primary HCB predictors (including tree, stand, stand spatial structure, density and competition factors) and their relationships with the HCB of the four tree species groups. The relative importance was higher for individual tree variables (77.54 %, 31.02 %, 31.12 %, and 73.69 % for coniferous, spruce-fir, hard broadleaved, and soft broadleaved groups, respectively) and stand variables (5.00 %, 20.34 %, 11.03 %, and 8.71 % for coniferous, spruce-fir, hard broadleaved, and soft broadleaved groups, respectively) compared with stand spatial structure variables (4.57 %, 12.14 %, 21.91 %, and 5.89 % for coniferous, spruce-fir, hard broadleaved, and soft broadleaved groups, respectively), density indexes variables (2.17 %, 1.28 %, 4.05 %, and 2.87 % for coniferous, spruce-fir, hard broadleaved, and soft broadleaved groups, respectively), and tree species variables (10.79 %, 35.20 %, 31.90 %, and 8.84 % for coniferous, spruce-fir, hard broadleaved, and soft broadleaved groups, respectively). BRT and Cubist were the best approaches for modelling the four species-group specific HCBs. Although spatial structure variables had minor relative importance, further in-depth investigations are warranted.