Quantile regression: prediction of growth and yield for a eucalyptus plantation in northeast Brazil

Regression analysis is a traditional technique to fit equations and predict tree and forest attributes. However, problems may occur when the data show high dispersion around the mean of the regressed variable, limiting the use of traditional methods such as the Ordinary Least Squares (OLS) estimator. Hence, the objectives were to propose a Quantile Regression (QR) methodology to predict tree growth and yield of a forest plantation without using the site index and compare it with the predictive accuracy of the Clutter model. The data came from clonal plantations of Eucalyptus grandis x E. urophylla located in the north of Bahia state (Brazil), and age ranged from 20 to 89-months-old. The set of permanent plots was  divided into 50% for model training and the remaining for validation. The volume prediction methods, Clutter and QR, were evaluated using scatter plots of the relative error and graphs of the observed versus predicted volumes, as well as by the following statistics: mean prediction bias in percentage (BIAS %) and root mean square error in percentage (RMSE %). The volume predictions by QR resulted in higher accuracy compared to those obtained by the Clutter model. The QR demonstrated efficiency to predict volumes for different ages of eucalyptus plantations and can also be adapted to other forest areas and species worldwide.