Unveiling the chlorine demand and decay in aged drinking water supply systems: Complementing experimental results with modelling and predictive artificial intelligence

Waterborne diseases claim over a million lives year-round, with diarrhea being a leading cause of death for children under five in the Global South. An astonishing 2.2 billion people, nearly a quarter of the global population, lack access to safe drinking water, while in the Global South even tap water can be unreliable and unsafe. This highlights the need for robust chlorine demand and decay assays in these settings. Experimental studies are indispensable, but given that numerous factors (e.g., temperature, pH, and water quality) are at play, often interacting in a non-linear fashion, they should be complemented by predictive modelling, preferably using machine learning (ML) algorithms. Here, an artificial intelligence (AI)-based ML model, informed by the parallel-first-order chlorine decay model (PFOM), was developed to predict chlorine consumption and decay within an ageing drinking water distribution network in South Africa. The ML model quantified the covariance (R2 = 0.0002–0.08) between chlorine decay parameters and successfully determined the optimal neural network configuration to fit the experimental data. Modelling of chlorine decay using ML has simplified and saved on the complexity of chlorine decay models. Chlorine demand was linearly dependent on the chlorine dose (R2 = 0.95), with the optimum chlorine dose being 6.30 mg/L, reflecting numerous chlorine-consuming substances in the treated water and also suggesting problems with the formation of chlorine disinfection by-products (DBPs), making ML use indispensable. Hence, the ML model should be further coupled with solute transport models to simulate the effect of distance using linear regression-based chlorine decay models to correlate the chlorine dosage with model parameters.