ESTIMATING THE SAMPLING ERROR FROM THE COVARIOGRAM OF SPATIALLY CORRELATED DATA

Estimando erro amostral a partir de covariogramas de dados espacialmente correlacionados

Abstract

The total sampling-error attached to a set of samples has a central role in the selection of the statistical method to extract information from this noisy data. However, commonly direct measurements of the sampling error are not available and then, the magnitude of the error is unknown. In this article, we present a mathematically sound solution for estimating the sampling error directly from spatially correlated observations. The method is based on the difference between the global variance and the inferred y-axis intercept of the covariogram computed from the same data. We developed the mathematical proofs of the method, and its performance is analyzed by applying it to five variables from a stream-sediments dataset of a multi-element geochemical survey. The estimated total sampling error is satisfactory close to the value experimentally measured by field replicates.