Assessment and prediction of irrigation water quality using machine learning techniques: a case study from the Bouhamdane Basin, Guelma Region, Eastern Algeria

Abstract

This study evaluates the performance of machine learning (ML) algorithms in predicting the Irrigation Water Quality Index (IWQI) for irrigation in the Bouhamdane Basin, with XGBoost ML-2 emerging as the most accurate model (R2 = 0.93, root mean square error (RMSE) = 0.047). Other boosting-based methods like ADABoost ML-4 (R2 = 0.79, RMSE = 0.085) and CatBoost ML-2 (R2 = 0.87, RMSE = 0.065), as well as random forest (RF) ML-2 (R2 = 0.89, RMSE = 0.06), also demonstrate strong predictive capabilities, outperforming linear and support vector regression models. Additionally, voting ensemble models (R2 = 0.90, RMSE = 0.06) highlight the effectiveness of integrating multiple algorithms. The study also analyzes key water quality parameters, revealing variability in irrigation suitability, with some samples posing salinity and sodicity risks (sodium adsorption ratio (SAR) up to 24.509, IWQI as low as 16.86) and others being suitable (IWQI up to 53.80). Overall, the study underscores the significant role of advanced ML techniques in improving water quality assessment for irrigation, offering valuable insights for environmental monitoring and agricultural water management.