Mixed-effect modelling of Tectona grandis L.f. yield at stand-level

Abstract

The prediction of volume from commercial plantations is essential to manage Tectona grandis L.f. stands. However, the assumptions of regression, especially homoscedasticity and residual independence, are often ignored when fitting longitudinal regression models. Thus, this work aimed to build a mixed-effect model to obtain accurate estimates and correct inferences on the volume yield of Tectona grandis stands. A descriptive analysis of stand structure variables obtained in 46 permanent plots was performed, in which the covariates for a yield model with the highest correlation with volume were selected. Multicollinearity was measured by the corrected variance inflation factor, selecting covariates with values close to 10. To incorporate the asymptotic characteristic in the prediction of total volume, the logistic function was modified to include the selected covariates. Then, mixed-effect models were fitted by the restricted maximum likelihood method, where the plot was included as a random effect. The covariates basal area, logarithm of the quadratic mean diameter, and age were selected to compose a volume model for Tectona grandis. Incorporating plots as random-effects on the basal area of covariates’ relationship provided the best goodness-of-fit and satisfied the regression assumptions for a yield model. These results aid in the management of Tectona grandis by providing accurate volume yield estimates for decision-making, optimal rotation, and resource allocation.