Agrometeorologia e produção dos capins massai e tanzânia nas regiões amazônica e semiárida do Brasil: uma abordagem estatística.

Abstract

Forage production results from the systemic action of environmental factors that act on pasture ecosystems. The study of the biophysical processes involved and the decision making based on predictive models, must take into account the conditions of the environment in which the plant grows. In the search to understand how environmental variables act in the production of forage, and to obtain models applied to equatorial regions, the objective in this study was to evaluate the biomass production of the cultivars Guinea and Massai, in studies carried out under the influence of both the Amazon biome and the Brazilian semiarid climate, from the perspective of multivariate analysis and generation of empirical models. Data from six experiments: two with Guinea grass and one with Massai grass, carried out in Igarapé-Açu - PA; and two other experiments: two with Tanzânia grass carried out in Pentecoste - CE and the other with Massai grass in Fortaleza - CE; were subjected to analyses of canonical correlation (CCA), discriminant function (DFA), and principal components (PCA). Records of accumulation of total forage (TFAc), leaf blade (LBAc), stems (SAc), tiler population density, and canopy height were used as agronomical variables; and a set of agrometeorological variables. Data from experiments carried out in the Amazon region were used to generate models using machine learning methods, subsequently validated with data from the semiarid region. The TFAc was considered the response variable, whereas the agrometeorological variables were considered the independent variables. Multivariate statistical methods were able to characterize the experiments from the perspective of the variability of environmental conditions and cultivars studied. This characterization demonstrated a greater variability of the experiments carried out in the north region than those carried out in the northeast region. The canonical correlation analysis can correlate well with the agrometeorological variables and productive variables. And the experiments were separated by two discriminate functions, one that discriminated the experiments in terms of regions and the other in terms of cultivars. As for forage prediction modeling, machine learning methods, particularly the partial least squares regression method, were able to generate prediction models from data from the Amazon region to predict forage biomass production in the semiarid region.