Prediction of irrigation water quality parameters using machine learning models in a semi-arid environment

Abstract

Evaluation of the water suitability for irrigation purposes using conventional approaches is generally expensive because it requires several parameters, particularly in developing countries. Therefore, developing accurate and reliable models may be valuable to overcome this issue in the management of the water used in agriculture. To achieve this purpose, 8 Machine Learning (ML) models namely: Artificial Neural Network (ANN), Multiple Linear Regression (MLR), Decision Tree, Random Forest (RF), Support Vector Regression (SVR), k-Nearest Neighbour (kNN), Stochastic Gradient Descent (SGD) and Adaptive Boosting (AdaBoost) have been developed and validated for predicting of 10 Irrigation Water Quality (IWQ) parameters such as Sodium absorption ratio (SAR), adjusted SARa, Exchangeable Sodium Percentage (ESP), percentage of Sodium (%Na), Residual Sodium Carbonate (RSC), Permeability Index (PI), Kelly Ratio (KR), Chloride Cl−, Magnesium Absorption Ratio (MAR), and TDS dissolved in water surface of Bouregreg watershed in Morocco using electrical conductivity (EC) and pH as input variables. 300 samples are analysed at 9 monitoring stations across four main rivers, processed and selected to train and validate the models. The results have revealed that, except for SVR and k-NN models and MAR and PI parameters, all other models are highly accurate in predicting the other parameters with coefficients of correlations (r) with ranges of [0.56, 0.99], and [0.64, 0.99] for training and validation processes sequentially. Furthermore, this study attempts to generalize the 6 ML models developed and validated to the Cherrate and Nfifikh watersheds that are different from Bouregreg watershed. The results of the generalization attempt have shown that the ML models are fairly generalized for TDS, SAR, and SARa parameters to Cherrate watershed and for TDS, chloride, ESP and %Na parameter for Nfifikh watershed. The results of this study have also demonstrated that the machine learning models are efficient tools for accurately predicting the quality of irrigation water by only using the parameters that can be directly measured in a short time. Consequently, the implementation of the automated sensor technologies coupled with ML models improve the control of water quality and will help the farmers to manage the irrigation water quality.

Keywords

• Sodium absorption ratio
• Irrigation water quality
• Farmers
• Prediction accuracy