Accuracy of tree profile estimation in planted forests

Résumé

Theoretically, the tree shape is composed of three geometric segments: a neiloid at the base, a paraboloid in the middle, and a cone at the top. Modeling of this form aims to estimate the allocation of trees to different wood products, such as sawtimber, roundwood, pulp logs, firewood, or charcoal. Taper models can be categorized as non-segmented, segmented, sigmoidal, form exponents, and, more recently, adjusted by artificial neural networks (ANNs). Most studies have focused on selecting the best model by evaluating its performance under homogeneous conditions, such as single species, clone, spacing, and age. This study tested the hypothesis that within variations among trees under the same experimental setup can produce modeling errors that are as significant as those resulting from differences among trees across distinct conditions. To this end, data from various species, clones, ages, and spacing under a wide range of edaphic and climatic conditions were integrated. Different taper models were applied and their errors were compared using the most common performance metrics in the literature: the percentage residual standard error (Syx%) and, when unavailable, the root mean square error (RMSE). The results showed that modeling with integrated datasets produced intermediate error values compared with those reported in other studies. This suggests that, even when spacing, age, and environmental conditions are held constant, shape variation among trees of the same species or clones limits gains in precision, even when using models better suited to specific stem profiles.