Validation first removes part of the data(test dataset) and then uses the rest of the data(training dataset) to develop the trend and autocorrelation models to be used for prediction. If the prediction works for the test dataset(choice of semivariogram model, choice of lag size, choice of search neighborhood, etc), we can feel comfortable that it also works for the whole dataset.
Note: In the process of prediction, it is better for the whole sample dataset to be divided into two parts; one part with majority of the data is used for prediction and the other for verification.