Vol.:(0123456789) 1 3 Environmental Earth Sciences (2020) 79:456 https://doi.org/10.1007/s12665-020-09204-y THEMATIC ISSUE Predictive modeling for U and Th concentrations in mineral and thermal waters, Serbia Marina Ćuk Đurović 1  · Igor Jemcov 1  · Maja Todorović 1  · Ana Mladenović 1  · Petar Papić 1  · Jana Štrbački 1 Received: 30 June 2019 / Accepted: 12 September 2020 © Springer-Verlag GmbH Germany, part of Springer Nature 2020 Abstract The objective of this paper was to determine background values (BV) and anomalous values (AV) of U and Th in ground- water and to establish hydrogeochemical conditions which lead to the elevated concentrations of these elements in ground- water. The methodology included planning and collecting of water samples, laboratory work, and assessment of BV and AV concentrations in accordance with the dataset distribution, based on consideration of hydrogeochemical conditions in the hydrogeological system. Groundwater sampling included 144 occurrences of mineral and thermal water from Serbian territory, belonging to diferent hydrogeological systems. Field parameters were measured for temperature (T), pH, electri- cal conductivity (EC), oxidation–reduction potential (ORP), dissolved oxygen (DO), and carbon dioxide (CO 2 ). Standard laboratory measurements were applied for the determination of major chemical components (Ca, Mg, Na, K, Cl, HCO 3 , and SO 4 ) and U and Th concentrations were measured by ICP-MS. The frst step for obtaining U and Th threshold values was based on non-parametric statistical analysis on the data sets. Further analysis of threshold values enabled establishing hydro- geochemical conditions infuencing elevated concentrations of U and Th and setting up the logistic regression (LR) model. Diferences in the hydrochemical properties of U and Th can be observed based on predictor variables from LR models. Physico-chemical parameters Eh and pH, groundwater type, and geochemical environment (cretaceous igneous rocks) were signifcant predictors for elevated uranium concentrations, while signifcant predictors in the thorium LR model were the pH value, the concentration of SO 4 in the solution, and the water-bearing rocks (tertiary igneous rocks). Keywords Radioactive elements · Groundwater · Hydrogeological system · Non-parametric statistical analysis · Logistic regression model Introduction Predictive models have been increasingly used to generate predictions across various disciplines in the environmental sciences (Li 2017). The importance of applying predictive models in environmental studies is often related to manage- ment issues, such as risk management and spatial planning solutions (Chow et al. 2019). Interest in the use of Predictive models for regulatory purposes has also been growing, and many models have been evaluated under the diferent guid- ance and acceptability criteria (Tunkel et al. 2005), wherein the logistic regressions presents one of the most common types of predictive models. Large-scale predictions created by incorporating data collected from diferent regions into models make an opportunity for developing conceptual regional models that will contribute to the understanding of natural processes, making environmental studies to become a leading factor in the development of a sustainable society. This article is a part of the Topical Collectionin Environmental Earth Sciences on “Mineral and Thermal Waters"guest edited by Drs. Adam Porowski, Nina Rman and Istvan Forizs,with James LaMoreaux as the Editor-in-Chief. Electronic supplementary material The online version of this article (https://doi.org/10.1007/s12665-020-09204-y) contains supplementary material, which is available to authorized users. * Marina Ćuk Đurović marina.cuk@rgf.bg.ac.rs * Ana Mladenović ana.mladenovic@rgf.bg.ac.rs 1 University of Belgrade, Faculty of Mining and Geology, Belgrade, Serbia