Comparison of spatiotemporal variogram functions based on a sparse dataset of groundwater level variationsComparison of spatiotemporal variogram functions based on a sparse dataset of groundwater level variations Peer-Reviewed Journal Publication Δημοσίευση σε Περιοδικό με Κριτές 2020-04-022019enWe conduct a spatiotemporal geostatistical analysis of groundwater level data using space–time Residual Kriging (STRK). We compare the performance of three spatiotemporal variogram models in mapping groundwater level data. All models involve the same space–time trend function; the first model is a separable product variogram that involves Matérn space and time functions, while the second and third use non-separable functions. The non-separable models include a diffusive space–time variogram and a spatiotemporal extension of the Spartan variogram. The non-separable models are applied for the first time to groundwater level data. Our analysis shows that the Spartan and the non-separable diffusive variograms fit the experimental variogram of the residuals better than the separable Matérn model and improve the cross-validation results. The Spartan model provides the best cross validation measures among the three models. We also generate maps of groundwater level which illustrate the spatiotemporal variability in the basin.http://creativecommons.org/licenses/by/4.0/Spatial Statistics34December 2019 Varouchakis Emmanouil Βαρουχακης Εμμανουηλ Christopoulos Dionysios Χριστοπουλος Διονυσιος Elsevier Matérn Non-separable variogram Space–time Kriging Space–time trend Sparsely gauged basin Spartan model