Estimadores de semivariancia: analise de desempenho no mapeamento da precipitação anual

Abstract

The process of spatial interpolation is of great importance for inference of hydrological variables in places where information is not available. Among the various methods of spatial interpolation, the kriging family stands out, this is a geostatistical, non-biasedspatial interpolation process of minimum variance, in which the classic experimental semivariogram of Matheron is commonly used. Although this estimator is the most used, there are already other estimators of experimental semivariance more efficient than Matheron, since, due to the quadratic difference in its formulation, it is sensitive to atypical points in the data set, thus compromising the result of the interpolation. Considering these assertions, the present work was developed with the objective of evaluating the spatial interpolation performance of the mean annual total precipitation of the state of Minas Gerais, considering the period 1990-2015, using ordinary kriging from different semivariance estimatorsexperimental. The classical estimator of Matheron-(MTH), Cressie Harkins (CH), Cressie Medians (MED) and New-1 was used. As a result, it was verified, through the paired Wilcoxon test, that the kriging obtained by the evaluated estimators are statistically different at the significance level of 1%, and these differences could also be noticed visually on the maps. The experimental semivariance estimator New-1 presented the best performance