Detecção e influência de outliers na qualidade de modelos de relação hipsométrica sob o ponto de vista preditivo

Abstract

Models that express the relation between tree height and tree diameter are called hypsometric relationship models, where the objective is to estimate the height of unmeasured trees, reducing expenses in forest inventories. Among the sample used to adjust the model of hypsometric relation is common to observe the presence of outliers, where these can provide unstable estimates in the parameters of the adjusted model, as well as provide a worse performance in predicting heights. The objective of this study was to evaluate the influence of outliers on hypsometric models and to compare different approaches as alternatives in the improvement of these models against the presence of atypical values. Tectona grandis data were obtained from commercial plantations of Floresteca, where from these, hypsometric relation models were adjusted through the MQA, MQO with the accommodation of the atypical values and the readjustment of these models with subsequent identification and treatment of the outliers. The model selected by the AIC, relative RMSE and residual distribution was the Embrapa model, in which it was used to evaluate the performance of the approaches proposed in this work, where the performance between them had a negligible difference in relation to the adjustment of the model through the OLS With accommodation of the atypical values.