Confronto del codice di diversi pacchetti software per risalire alle cause delle differenze nei risultati: Ordinary kriging in ambiente FOSS

Nowadays, computers are the only tool allowing to complete complex procedures in reasonable time. Many scientists rely on dedicated software, citing it in their papers, to ensure reproducibility of the experiments. However, reproducibility alone does not certify that the software produces correct results. The risk of running an inaccurate computer procedure increases with the complexity of the mathematical/statistical issue that it addresses. The ability of a computer language to translate formulas depends on a number of factors, including the availability of well-tested libraries and the skills of the programmers both in the programming language and in the addressed problem. In geostatistics, kriging is a widely used interpolation technique. Many software packages, based on different programming languages, are able to perform kriging calculation, but few of them ensure an up-to-date development and an open access to the source code. Since kriging represents a probabilistic approach providing for “estimates” of a variable over space, differences among different packages are expected. Nevertheless, results should be consistent among them so that the choice of a package should not affect resulting estimates about the variable being considered. The implementations of ordinary kriging in the commonly used Free and Open Source software packages, gstat and geoR (running under R software) and the Python package hpgl are suitable for an exhaustive comparison, because the derived maps achieved by different packages can always be compared but only accessing the code allows to find the root of the differences in the outputs. The comparison procedure being proposed could represent a model for further geoscience software comparisons, in order to promote software choice based on quantitative and exhaustive evaluation of their performance.