Chemometrics of karst systems: Monitoring climate impacts on groundwater quality in Garadna Spring, Northern Hungary, using self-organizing maps and wavelet transform analysis

Monitoring the temporal evolution of groundwater quality in karst aquifers is vital, as these systems serve as major freshwater resources while being highly vulnerable to contamination due to rapid infiltration. However, obtaining continuous long-term datasets is often hindered by irregular sampling and limited technical resources, leading to data gaps that constrain comprehensive analyses. This study investigates groundwater quality dynamics in the Garadna Spring, Northern Hungary, by integrating machine learning, hydrochemical analysis, and wavelet transform techniques. A random forest (RF) model was used to impute missing anion concentrations (Cl, SO₄, and HCO₃), achieving performance in validation stage with an average coefficient of determination (R²) and mean square error (MSE) of 0.8 and 2.5 mg/L, respectively. The spring water is predominantly of the Ca–Mg–HCO₃ type, reflecting carbonate dissolution under near-equilibrium to oversaturated conditions for calcite and dolomite. Using self-organizing maps (SOMs), three major hydrochemical clusters were identified: (1) baseline high-mineralization conditions, (2) transitional phases, and (3) dilution effects following precipitation events. Cross-wavelet transform (XWT) analysis revealed that precipitation increases water levels rapidly with less than a 1-month lag, while its effect on total dissolved solids (TDS) is delayed by approximately 3–5 months. A strong water-level–TDS coherence at a 10-month timescale indicates prolonged water–rock interaction. The integrated approach demonstrates that hydrological responses in the karst system are faster than geochemical changes. These findings provide actionable insights for water resource managers seeking to monitor and predict groundwater quality in data-limited karst systems, supporting sustainable water management under climate variability.