Introduction
This study presents a robust geostatistical framework to evaluate the groundwater system in Southwest Thessaly, Greece. In the absence of dynamic simulation models, this methodology currently provides the optimal approach for a comprehensive quantitative and qualitative aquifer assessment, focusing on groundwater levels alongside critical physicochemical parameters: nitrates (NO3-), iron (Fe), and manganese (Mn).
Methods
Borehole data from 2013–2023 (encompassing recording months from February to December) were analyzed, with datasets ranging from 8 to 18 measurements per parameter and date. Targeting qualitative parameters with regulatory exceedances >10%, the Madlog method was applied for robust outlier exclusion. Spatial structure was determined using global spatial autocorrelation indices (Moran’s I and Geary’s C). Subsequently, deterministic and geostatistical interpolation methods—including Radial Basis Functions (RBF), RBF/Spline, Ordinary Kriging (OK), Natural Neighbor, Inverse Distance Weighting (IDW), and Empirical Bayesian Kriging (EBK)—were evaluated using Geographic Information Systems (GIS). Method selection was strictly guided by parameter-specific literature, sample size constraints, and spatial behavior classifications derived from the indices.
Results
Autocorrelation indices successfully quantified organized spatial behavior, though results and spatial interpolation efficacy exhibited significant spatiotemporal variability across dates. Rigorous cross-validation demonstrated that optimal interpolation accuracy is not universal but intrinsically dataset-dependent. IDW yielded the highest predictive accuracy for NO3- (RMSE = 9.1961) and Mn (RMSE = 20.4387). Conversely, OK performed optimally for Fe (RMSE = 49.7947), while RBF was the most accurate estimator for groundwater levels (RMSE = 2.6507).
Conclusions
Relying exclusively on the visual quality of interpolated maps is insufficient for an accurate hydrogeological assessment. By synergistically integrating Madlog outlier exclusion, spatial autocorrelation, and objective cross-validation, this study establishes a highly reliable spatial mapping framework, delivering a scientifically justified evaluation of the aquifer's spatiotemporal behavior.