Simulação de fatores que afetam as predições obtidas por Krigagem ordinária.

Abstract

The objective of this study is to identify how many factors can affect the variance of the predictions obtained by Ordinary Kriging. For this, different phenomena were simulated in order to obtain the predictions of Ordinary Kriging, using the three usual semivariogram models, different amounts of neighbors to perform the prediction, different spatial dependence indexes and sample size. It was observed that although all the models tended to produce non-biased predictions, the spherical model tended to show a greater variability of the prediction errors in most phenomena simulated, while the Gaussian got smaller. It was also found that when the value of spatial dependence indexes (FDI), the grid size and the number of nearest neighbors were increased, the models tend to have lower variability for all SDI studied, and these ones also tend to stabilize when it reaches a certain amount of neighbors. About the evaluation of the estimates, the exponential model tended to be the one with the worst rating on the variability of the predictions and the spherical the one with the best evaluation in most simulated phenomena. Nevertheless, when it increases the amount of neighbors, the models behavior changes according to the spatial dependence index.