Non-linear quantile regression in modeling the diametric growth of cedar (<i>Cedrela fissilis</i>)

Abstract

Understanding the dynamics of tree growth is extremely important to develop effective forest conservation and management strategies. Generally, tree growth is well fitted by non-linear regression models. However, it can commonly present problems caused by heteroscedasticity or possible asymmetry in the distribution of residues. An alternative to overcome this problem is quantile regression, which allows estimates at different quantiles, thus generating a more complete mapping of the development of the forest under study. The objective of this study was to compare the adjustment of the non-linear models Logístico, Gompertz, von Bertalanffy, Brody, Chapman-Richards and Weibull using the least squares method and quantile regression, for data on diameter at breast height (DBH) accumulated over the overtime for 56 trees sampled in native forest using a non-destructive technique. The coefficient of determination, the mean absolute deviation and the Akaike information criterion were used to evaluate the quality of the adjustments and the suitability of the models was verified through residual analysis, with the Brody model being the one that best adhered to the data. All computational analysis was carried out using the free software R.