Individual tree volume equations are essential tools for forest inventories, since wood content of individual trees can not be measured precisely, except by means of destructive methods.  In eucalyptus forests, the prediction of tree volumes is seldom done by the application of a single volume equations for the entire forest.  Volume equations are usually specific for genetic material (clones), properties and project/age of the forest.  In this study, we consider the influence of including the qualitative variables related to the spacial scale (municipalities, properties, projects and stands) in volume equation models and its impact in model selection.  A total of 11 linear models were fitted, including both standard volume equation models (DBH and height as predictors) and local volume equation models (only DBH as predictor).  Models were fitted by least squares and were compared using the coefficient of determination, residual standard error and several indexes based on model residuals.  The qualitative variables were fitted as dummy variables changing onlythe intercept regression coefficient and changing both the intercept and the slope coefficients of all predictors in the model.  The performance of all models was improved by adding the qualitative variables.  As expected, the improvement increased as the qualitative variable increased in spacial resolution from municipalities to properties, to project and to stands, but the difference in performance among the models decreased with spacial resolution.  The selection of the appropriate spatial resolution (qualitative variable) was more important to determine volume equation performance than the selection of the actual model.

03 - Qualitative Variables and the Selection of Volume Equation Models in Forest Inventory