As global water scarcity intensifies, particularly in arid and semi-arid regions, the reuse of treated wastewater (TWW) has emerged as a critical strategy to support sustainable agriculture. However, maintaining the quality of TWW, especially its Total Dissolved Solids (TDS) content, is essential to safeguard soil health and crop productivity. This study proposes an artificial intelligence (AI)-based predictive framework for accurately estimating TDS levels in TWW using multivariate water quality parameters collected from the irrigation zones in the Al-Hassa region of Saudi Arabia. The dataset includes measurements of various water quality parameters obtained through an IoT-enabled multiprobe monitoring system. Based on descriptive statistics and correlation analysis, three input configurations were formulated to train and evaluate four AI models, namely Support Vector Regression (SVR: M1, M2, and M3), Fine Gaussian Support Vector Machine (FGSVM: M1, M2, and M3), Stepwise Linear Regression (SLR: M1, M2, and M3), and Adaptive Neuro-Fuzzy Inference System (ANFIS: M1, M2, and M3), where M1, M2, and M3 correspond to the first, second, and third input configurations, respectively. Among these, the ANFIS-M2 model exhibited superior performance, achieving an R2 of 0.9635 during training and 0.9865 during testing, with a remarkably low RMSE of 0.0128 and MAE of 0.0059 on the testing dataset. On the other hand, FGSVM-M3 also delivered strong and consistent results, with R2 values of 0.9774 (training) and 0.9216 (testing), and a testing RMSE of 0.0309. The SLR models provided moderate accuracy and interpretability, and SVR models exhibited lower R2 values on the testing set, indicating limited generalization performance. These findings highlight the strength of AI models, particularly ANFIS, in capturing nonlinear relationships in the water quality data. It is therefore believed that this research demonstrates the potential of enhanced accuracy and efficiency in monitoring treated wastewater quality, enabling sustainable reuse and optimized water management through data-driven prediction of TDS.
Shah et al. (Mon,) studied this question.