Real-Time Water Quality Monitoring via Impedance Spectroscopy and Machine Learning

Water quality is crucial for plant growth, with factors like salinity levels significantly impacting crop health. Variations in water quality, especially from wells, necessitate regular monitoring. Existing methods are often costly, timeconsuming, or unsuitable for continuous monitoring. This study utilizes impedance spectroscopy for real-time water quality monitoring in irrigation systems and machine learning methods for prediction. The proposed method captures spectral features and employs a compact machine-learning model for efficient and accurate pattern recognition, outperforming traditional electric conductivity measurements. Experiments measured various spectral features from water with different concentrations of NaCl, MgSO4, and their mixtures, across 1kHz to 1MHz using an Analog Discovery 2 device. Data from these experiments were used to estimate solute concentrations. Machine learning methods, including Random Forest and Multilayer Perceptron, were employed to predict NaCl and MgSO4 concentrations in mixed samples. Results demonstrate significant improvements over existing methodologies, supporting continuous monitoring. With 55.4 ppm MAE for NaCl and 121.5 ppm MAE for MgSO4, the prediction results are promising for real-time water quality monitoring. Implementing farmer-specific devices could enhance agricultural automation by setting thresholds, warnings, and enabling automatic