Modelos estatísticos e técnicas de inteligência artificial para estimativa do volume de clones de Eucalyptus spp. com adição de variáveis climáticas

Abstract

The Gypsum Pole of Araripe is the main plaster producer in Brazil. To dehydrate the gypsum mineral that becomes plaster, the main energy source is firewood from the Caatinga, which has been exploited generally in an ilegal manner. Nowadays the effects of deforestation in the Caatinga combined with climate change have made the supply of energy resources scarce. As a way of mitigating the effect of deforestation and reducing the suppression of Caatinga vegetation, experiments with fast-growing forests were implanted in the region. In order to provide a more adequate development of the management of these forests, it is necessary to adjust local statistical models that are more efficient and that take into account changes in climate variables. Thus, the objective of this research was to adjust nonlinear models and machine learning algorithms and to evaluate whether the inclusion of climatic variables generates equations with greater precision and generalization power in the estimation of production and volumetric growth of textit Eucalyptus in a semi-arid climate region in Northeast Brazil. The data were obtained through a research carried out at the IPA Experimental Station, municipality of Araripina - PE, with three clones of Eucalyptus spp. (C39, C41, C11) planted in spacing 3 mx 3 m, 4 m x 2 m, 3 m x 2 m, 2 m x 2 m and 2 m, x 1 m, in a multivariate experiment of repeated measures, implemented in 2010. A height and diameter measurements were taken every six months. At the age of 96 months, the volume was determined by the Smalian method. The rainy season database (2002 to 2009) was obtained from an experiment developed in the same region with the same silvicultural treatments. Nonlinear models and artificial intelligence algorithms were used for volume production, growth and prognosis. The adjustments were evaluated based on the corrected Schlaegel adjustment index (IA c), Furnival index (IF), standard error of percentage estimate (Sxy %), root mean square error (RMSE), mean absolute error (MAE) and graphical analysis of the residues. Volumetric production differed among clones and between different levels of density. Symmetric models and generalized nonlinear models produced equations with high AIc values and low values of S xy % with residuals meeting the assumptions of the regression. The artificial intelligence techniques have efficiently adapted the estimate of production and growth in volume, generating statistics of good quality. The inclusion of climatic variables in the growth and prognosis models improves the estimates.