摘要：The accuracy of results obtained by geostatistical analysis, regarding spatial prediction, depends substantially on determining more e±cient sampling schemes, with a reduced number of samples. This in the sense that the obtained results are similar to the actual results in the area, thus reducing operational costs. By using simulated data, the present work has laid plans for e±cient spatial sampling in the prediction of variables with spa- tial dependence. The simulated annealing and hybrid genetic algorithm (GA) methods of optimization are used, considering the mean of the prediction variance as objective function. In addition, this work allows us to resize a sample conˉguration. This has already been applied to an experiment of precision agriculture in the cultivation of soy- beans in Paran.a, reducing by 50% its sample size and minimizing the e±ciency losses of spatial prediction that this sample reduction may cause. The results for the simulations show that the optimized sample conˉgurations produce lower estimates for the mean variance of the spatial prediction and better estimates for the characteristics related to spatial prediction in the studied area. Moreover, for the experiment considered in this study, the results show that the sample conˉguration reduce by hybrid GA shows a greater similarity with the initial sample conˉguration. Thus, the 50% reduction in the sample size by using the hybrid GA produces e.ective results for the classiˉcation of potassium (K) fertilizer in the area. Therefore, this reduced sample conˉguration may be used in future experiments for this area, reducing by 50% costs of chemical analysis of soil, without great loss of e±ciency in the conclusions drawn by the spatial prediction.
