Evaluation and prediction of groundwater quality for irrigation using regression and machine learning models

ABSTRACT This study evaluates and predicts six water quality indices such as sodium adsorption ratio (SAR), Kelly's ratio (KR), percentage sodium (%Na), permeability index (PI), exchangeable sodium percentage (ESP), and irrigation water quality index (IWQI) using multivariate regression models (MLR, PLSR, PCR, and WLSR) and machine learning (ML) algorithms (ANN, SVM, CART, CRRF, and KNN). The study analyzes data from 360 dug wells in Sundargarh district, India, during 2014–2021 with 70% used for training and 30% for testing. Spatial mapping of SAR, KR, ESP, and PI exhibits higher suitability of groundwater. The Mann–Kendall test of trend analysis shows a monotonic increasing and decreasing trend for SAR, KR, %Na, ESP, PI, and IWQI, respectively, at p > 0.05 during 2014–2021. Principal component analysis and discriminant analysis identify Na+, SAR, KR, %Na, and PI as the most influential WQ variables affecting the groundwater quality for this study area. MLR and WLSR models are superior in predicting SAR and ESP, while ANN is the best-suited ML model for SAR, KR, %Na, PI, and ESP. CRRF predicts IWQI with a relatively higher accuracy. These findings demonstrate the effectiveness of ML models in improving irrigation water quality assessment, providing valuable insights for groundwater-based crop management.