Comparison of prediction methods for oxygen-18 isotope composition in shallow groundwater

Groundwater is the most important source of drinking water in the world. Therefore, information on the quality and quantity is important, as is new information related to the characteristics of the aquifer and the recharge area. In the present study we focused on the isotope composition of oxygen (δ¹⁸O) in groundwater, which is a natural tracer and provides a better understanding of the water cycle, in terms of origin, dynamics and interaction. The groundwater δ¹⁸O at 83 locations over the entire Slovenian territory was studied. Each location was sampled twice during the period 2009-2011. Geostatistical tools (such us ordinary kriging, simple and multiple linear regressions, and artificial neural networks were used and compared to select the best tool. Measured values of δ¹⁸O in the groundwater were used as the dependent variable, while the spatial characteristics of the territory (elevation, distance from the sea and average annual precipitation) were used as independent variables. Based on validation data sets, the artificial neural network model proved to be the most suitable method for predicting δ¹⁸O in the groundwater, since it produced the smallest deviations from the real/measured values in groundwater.