Modelo de crescimento e produção de Clutter adicionado de uma variável latente para predição do volume em um plantio de Eucalyptus urograndis com variáveis correlacionadas espacialmente

Abstract

Growth and yield models are important to predict changes in a forest and provide information that helps to make more reliable decisions. Geo-statistical techniques used jointly with growth and yield models can be useful to obtain more precise estimates of yield when the data set is spatially correlated. Therefore, this paper is aimed to compare the Clutter model (1963) and the same model with a spatial component in terms of precision of volumetric estimates. The working hypothesis was that inserting a spatial component into Clutter's model would provide improvement on its fitting and predictive capacity when applied to a spatially dependent data set. The data set used was from a Eucalyptus plantation where 117 plots of 400 m2 each were measured from the 2nd to the 6th year of growth. The traditional model proposed by Clutter (1963) and the same model with a spatial component in the basal area equation were fitted with likelihood methods. The second model was tested with several correlation functions for the spatial component. The models were compared using the Akaike Information Criterion (AIC) and the Residual Standard Error (RSE). The results showed that the model with a spatial component fitted better than the traditional model to the data set. There was a gain of precision when predicting basal area and volume using the model with a spatial component with the spherical, exponential, Mátern (k=0.2 and k=0.3) and powered exponential correlation functions. The volume predictions made by the model with a spatial component reduced the residual standard error to 6.57%.