Spatiotemporal evaluation of irrigation groundwater quality in Hungarian agricultural sites using hydrochemical and machine learning approaches

Abstract Groundwater is a critical source for irrigation in many agricultural regions, particularly in Hungary where surface water is limited. This study investigates how groundwater quality has evolved over time and its implications for sustainable irrigation. The research integrates hydrochemical assessment with machine learning to enhance the spatiotemporal evaluation of irrigation water quality. Groundwater samples from the Debrecen area (2019–2024) were analyzed using Hierarchical Cluster Analysis (HCA), irrigation indices (SAR, Na%, IWQI), and Self-Organizing Maps (SOMs) to understand spatial and temporal patterns. A Convolutional Neural Network (CNN) model was developed to predict IWQI from key water quality parameters, aiming to reduce manual calculation errors. HCA indicated low to moderate mineralization in most samples, while SOMs revealed notable spatial and temporal shifts, including gradual degradation due to natural and anthropogenic factors. IWQI assessments confirmed general suitability for irrigation, although localized risks due to salinity and sodium hazards were identified. The CNN model achieved high predictive accuracy (R² >0.97), streamlining IWQI estimation and minimizing human error. The findings highlight the utility of machine learning in groundwater quality monitoring and support more sustainable, adaptive irrigation practices to protect long-term soil productivity and food security.