A comparison of fixed- and mixed-effects modeling in tree growth and yield prediction of an indigenous neotropical species (Centrolobium tomentosum) in a plantation system

Abstract

Centrolobium tomentosum is a multipurpose pioneer tree species, indigenous in tropical South America and suitable for forest restoration, agroforestry and plantation systems. Despite its economic and ecological interest, no growth and yield models have been developed for this species so far. Fixed- and mixed-effects modeling can be used in model fitting, each technique having its pros and cons. Marginal predictions can be computed from fixed-effects models or randomized mixed-effects models. In forestry practice, models are seldom calibrated and mixed-effects models are mostly used to provide conditional predictions using only the fixed parameters, assuming that the random effects are zero. This study developed the first set of individual-tree growth and yield models for C. tomentosum and, by using the models, assessed the performance of three prediction approaches: fixed-effects models, conditional predictions of mixed-effects-models and marginal predictions of mixed-effects models. The fitted models predict maximum mean annual bole volume increments of 5.6–16.6m3/ha and optimal rotation lengths ranging from 11 to 21years, depending on site quality. Fixed-effects modeling was the best approach in growth and yield prediction, followed by conditional predictions of mixed-effects models, whereas marginal predictions based on mixed-effects models were in general the least accurate. Fixed-effects models should therefore be preferred in the absence of calibration data. However, since calibration is sometimes a feasible option, research articles should report both fixed- and mixed-effects models in order to enable the computation of the best predictions with and without the possibility of model calibration.