MIXED-EFFECT MODELS FOR VOLUMETRIC ESTIMATION OF LUMBER FROM NATIVE AMAZONIAN SPECIES

Abstract

The objective of the present work is to test fixed effect models and compare them with different configurations of mixed models to predict lumber volume. The database came from a medium-sized sawmill located in the rural area of the municipality of Porto Grande, Amapá, Brazil. A total of 50 logs of 9 native species were used, with their volume being measured by the Smalian method. A fixed effect model proposed by Lima et al. (2018) was fitted andcompared to mixed linear models with species as a random effect. The fitting of the fixed model and the mixed models were performed using the maximum restricted likelihood method. Akaike information criteria (AIC), correlation coefficient, Maximum Likelihood Ratio Test, Square Root of the Mean Square Error (RMSE), absolute mean distance and bias were used to select the best model. According to the established criteria, the mixed models obtained better fits when compared to the fixed model, providing a reduction in the RMSE of 0.3 for an average between the mixed models of 0.26. According to the results achieved, the mixed linear models were more efficient and accurate for volume estimation. The M7 mixed model was the best fit model. The use of mixed models enables greater accuracy in estimating volume.