Use of nonparametric regression methods for developing a local stem form model

Abstract

A local mean stem curve of spruce was represented using regression splines. Abilities of smoothing spline and P-spline to model the mean stem curve were evaluated using data of 85 carefully measured stems of Norway spruce. For both techniques the optimal amount of smoothing was investigated in dependence on the number of training stems using a cross-validation method. Representatives of main groups of parametric models - single models, segmented models and models with variable coefficient - were compared with spline models using five statistic criteria. Both regression splines performed comparably or better as all representatives of parametric models independently of the numbers of stems used as training data.