Prediction of irrigation water quality indices based on machine learning and regression models

Abstract

AbstractAssessing irrigation water quality is one of the most critical challenges in improving water resource management strategies. The objective of this work was to predict the irrigation water quality index of the Bahr El-Baqr, Egypt, based on non-expensive approaches that requires simple parameters. To achieve this goal, three artificial intelligence (AI) models (Support vector machine, SVM; extreme gradient boosting, XGB; Random Forest, RF) and four multiple regression models (Stepwise Regression, SW; Principal Components Regression, PCR; Partial least squares regression, PLS; Ordinary least squares regression, OLS) were applied and validated for predicting six irrigation water quality criteria (soluble sodium percentage, SSP; sodium adsorption ratio, SAR; residual sodium carbonate, RSC; potential of salinity, PS; permeability index, PI; Kelly’s ratio, KR). Electrical conductivity (EC), sodium (Na+), calcium (Ca2+) and bicarbonate (HCO3−) were used as input exploratory variables for the models. The results indicated the water source is not suitable for irrigation without treatment. A good soil drainage system and salinity control measures are required to avoid salt accumulation within the soil. Based on the performance statistics of the root mean square error (RMSE) and the scatter index (SI), SW emerged as the best (0.21% and 0.03%) followed by PCR and PLS with RMSE 0.22% and 0.21% for SAR, respectively. Based on the classification of the SI, all models applied having values less than 0.1 indicate good prediction performance for all the indices except RSC. These results highlight potential of using multiple regressions and the developed machine learning methods in predicting the index of irrigation water quality, and can be rapid decision tools for modelling irrigation water quality.