Application of multiple nonlinear regression models for short-term forecasts of karst aquifer water quality parameters: the example of Banja spring near Valjevo, Republic of Serbia

Karst spring waters, known for their high quality, represent an important resource for water supply. Since these waters are well-protected from external influences, they can usually be used directly with minimal treatment. To monitor the regime of these waters, it is necessary to establish continuous monitoring, which is often impossible due to the inaccessibility and uninhabited nature of karst massifs. The Banja spring (Petnica spring) represents one of the few springs in Serbia where daily monitoring of qualitative and quantitative parameters has been established. It is located near the Petnica Science Center (PSC), close to the city of Valjevo (Western Serbia). Daily measurements and sampling were conducted at this spring between 1991 and 2000, covering a total of 14 quality parameters as well as the volume of discharged water. To assess the correlation between precipitation and water discharge, rainfall amounts were measured using a rain gauge installed at PSC. This study presents results for the period from 1991 to 1995, focusing on discharged water (Q), pH value, electrical conductivity (Ec), total number of bacteria (TNB), turbidity (T), suspended sediment (SS), and daily rainfall sums (P). Auto-regressive (AR), cross-regressive (CR), and combined auto-cross-regressive (ARCR) simulation models were applied to the defined time series of qualitative and quantitative parameters. These models hold significant practical value in hydrogeological research and can be utilized for short-term forecasts of the analyzed parameters. Based on the obtained results, future studies should consider values of effective precipitation to develop more reliable models and achieve more accurate predictions.